Note: Descriptions are shown in the official language in which they were submitted.
CA 02925275 2016-03-23
Apparatus and Method of Animal Recognition using Nose Patterns
Technical Field
The present invention relates to an apparatus and method of an animal
recognition using
nose patterns, and particularly to an apparatus and method of an animal
recognition comprising a
body stabilizer unit to minimize movement of the subject animal for optimal
nose pattern image
acquisition, an image acquisition unit to obtain and store the nose pattern
image, and an image
recognition unit to generate, enroll, verify, and identify the raw or
processed nose pattern code
from the acquired nose pattern image.
Background Art
Animal identification has been around for thousands of years as indicated in
the Code of
Hammurabi dating back to about 1754 BC. At the time, the preferred method of
marking a body
part (branding) was used primarily to prevent the theft of valuable animal
assets like horses.
Nowadays, in addition to theft prevention and general proof of ownership,
animal identification
serves an important role in the production management and disease control of
livestock, the
management of endangered and protected species, as well as the essential
inspection process of
animal imports and exports. Globalization has increased the worldwide demand
for animals for a
variety of purposes, ranging from meat consumption to collecting exotic pets.
Accordingly, animals are being mass-bred and exported, but this has resulted
in the spread
of epidemics like the mad cow disease, which had previously been limited to
specific farms or
regions. Therefore, each and every state including the UN has sought to employ
an effective and
reliable animal tracking and identification system to manage the risks
involved in the domestic
and international production, distribution, and transportation of animals.
More recently, various
studies are under progress in an attempt to improve the traditional methods
and systems by
incorporating new developments in information technology.
Conventional methods of animal (livestock) management include: ear notching,
primarily
for pigs; plastic and bar-coded ear tagging, primarily for cows and sheep;
number designation on
1
CA 02925275 2016-03-23
neck chains for cows; freeze branding numbers or letters using metal cooled
with liquid nitrogen,
dry ice, or alcohol; paint branding; and tattooing. These procedures require
needless or painful
modifications or attachments to the animal's body, putting animals and the
necessary
professional handlers in potential harm's way. Even when executed without
complications, these
external markers or labels can be difficult to identify in the dark, or become
damaged by the
physical activities of the animal or by human acts of vandalism.
The alternatives to the above methods largely fall under two categories:
electronic and
biometric identification. Electronic identification requires the use of an
electronic ear tag,
injectable transponder, or a ruminal blouse to contain and be scanned for the
desired information.
However, unintentional damage to or intentional tampering of the microchip or
antenna, as well
as the unavailability of an appropriate scanning device can make
identification impossible. Also,
some studies have found that the material surrounding the microchip and
antenna unit can cause
tumors or tissue necrosis in the animals, providing significant reasons for
concern among owners
of companion and livestock animals.
The second alternative, on the other hand, is more promising. Biometric
identification relies
on the intrinsic characteristics unique to individuals without the necessity
of invasive procedures
and, unlike the traditional methods or microchip identification, the biometric
data of an animal
cannot be doctored. Current ongoing studies are seeking to make progress in
animal iris and
retina imaging, DNA analysis, and nose pattern imaging. However, the first
three have not been
developed enough yet to be practically applicable in the field.
As such, some limited efforts were made in the late 20th century, when the
uniqueness of
individual nose patterns on certain livestock became widely accepted, to
obtain and compare
nose patterns on individual cows or sheep in the same way fingerprints were
initially collected:
by cleaning the surface of the nose followed by obtaining ink impressions.
However, this method
is rather primitive and has consistently presented many issues in terms of
practicality and
accuracy; depending on the inexperience of the administrator there were
unwanted ink bleeds or
distortions on the print due to uneven application of pressure, often
resulting in disparate prints
of the same nose even when performed by a single individual. Also, a healthy
animal nose is
2
CA 02925275 2016-03-23
meant to maintain moisture through natural nasal irrigation or deliberate
licking, which means
each ink transfer is a laborious process.
Korean Laid-open Patent Publication No. 10-2004-0008072 presents the technical
configuration of portable information terminal for controlling cattle, while
Korea Laid-open
Patent Publication No. 10-2004-0006822 discusses a method of remote bovine
identification and
health monitoring using previously scanned nose pattern data via the internet
or a network .
However, due to their reliance on the traditional ink impression method before
scanning the
resulting ink print to obtain the nose pattern image, the limitations in
accuracy and the potential
for aberration arising from human error during the process are prevalent.
Moreover, the above
methods cater only to bovine subjects and thus are inapplicable to animals
with differently sized,
shaped and patterned noses.
Korea Laid-open Patent Publication No. 10-2002-00066771 presents the technical
configuration of a system of canine identification using nose pattern data
through a
communication network, but it does not specify the method of obtaining such
data.
On the other hand, US Patent Application No. 10/770,120 does disclose a
technical
construction of obtaining the nose pattern images of canine pets. The pattern
data are collected
by either macro shooting with a specialized camera to compensate for the
difficulty in focus
adjustments while manually holding the subject animal's muzzle, or by getting
an impression of
the nose -- similar to the traditional method -- using some pliable plastic
coated with ink or a
paper or card comprising two different layers of chemicals in place of ink.
With the macro shooting approach, size and spacing distortion in the nose
pattern image can
occur; and, as mentioned in the said patent as a point of concern, it is
difficult for an average
person operating a conventional digital camera or camcorder to make precise
focus adjustments.
Another method described in the above patent uses a polaroid camera where the
focus is set by
converging two irradiating beams, but it is highly likely that during the
process the light will
agitate and cause the canine subject to resist the restraint or move and
impede the photography.
Moreover, out in the field it is difficult to predict each subject animal's
sensitivity or reaction to
3
CA 02925275 2016-03-23
deliberate lighting and often smaller dogs are quite intimidated by the camera
itself, both adding
to the difficulty. Then there is also the problem of getting unwanted light
reflections off of the
moisture normally present on the nose skin when direct illumination is used
during image
capture.
The latter approach using contact impression is identical in its practice as
well as limitations
to the traditional inking method. In addition, a dog's tendency to actively
keep their noses wet
through licking when the moisture is deliberately wiped off means a hurried
effort is required to
obtain a clean impression, while the size and shape specification of the
equipment limits the
appropriate subjects to canine breeds.
Another prior invention related to the present one is the Japanese Laid-open
Patent
Publication 2003-346148, which prescribes that a glass frame or some other
transparent material
be pressed against a cow's nose to obtain the image with a digital camera for
analysis. However,
this approach is also similar to the traditional ink impression method wherein
a transparent frame
has merely replaced the decal paper, and presents the same distortion issues
that result from
direct contact with the nose as well as its limitation to bovine subjects.
Thus follows that there is a demand for a streamlined animal identification
system that does
not require professional operators, can overcome the aforementioned
encumbrances, and can be
easily and economically applied to small farm or livestock environments,
animal registration
systems, and even import and export inspections.
Disclosure of Invention
=
Technical Problem
An object of the present invention is the acquisition and recognition of
animal nose patterns
without making direct physical contact with the nose.
Another object of the present invention is the acquisition of nose pattern
images fit for
recognition by utilizing a body stabilizer unit that minimizes obstructive
movements in a subject
4
CA 02925275 2016-03-23
animal that behaves uncooperatively out of fear or aggression toward the image
acquisition
equipment or the operator, and maintains the ideal frontal capturing angle on
the subject's nose.
Yet another object of the present invention is the acquisition of good quality
nose pattern
images with the use of a body stabilizer unit designed to accommodate animals
of different sizes
and species.
Yet another object of the present invention is the acquisition of high quality
nose pattern
images by utilizing indirect illumination of appropriate wavelength regions
applied through a
light conduit subunit, light diffuser subunit, and spacer onto the subject's
nose to prevent
unwanted light reflections that may come off the layer of moisture on the nose
surface of subject
animals.
Yet another object of the present invention is the acquisition of high quality
nose pattern
images through the use of an image capture unit that is modifiable to
accommodate subject
animals of different species.
Yet another object of the present invention is to enable non-professional
users to acquire
nose pattern images fit for recognition with ease using the image acquisition
unit.
Yet another object of the present invention is to make possible the
identification of any
animal with a discernible nose pattern, regardless of species- or breed-
specific pattern types.
Yet another object of the present invention is to make identification possible
regardless of
the presence of extraneous physiological or environmental phenomena, such as
moisture, hair, or
dust of the subject animal's nose surface.
Yet another object of the present invention is to make identification possible
despite
reasonable variations in the image capturing angle.
CA 02925275 2016-03-23
Yet another object of the present invention is to generate a universal nose
code irrelevant
species or breed for use in identification.
Yet another object of the present invention is to use the most appropriate
method of
verification or identification for particular species or breeds.
Yet another object of the present invention is to increase the accuracy rate
of recognition for
each species of subject animals by comparing and matching previously stored
nose pattern
images to those newly obtained using the proper body stabilizer unit, image
acquisition unit and
image recognition unit.
Technical Solution
A technical solution of the present invention is to provide an animal
recognition apparatus
comprising a body stabilizer unit, image acquisition unit, and image
recognition unit.
Another technical solution of the present invention is to provide an animal
recognition
method comprising the following steps: selection of the appropriate body
stabilizer for the
species or breed of the subject animal, stabilization of the subject animal's
body using the
selected body stabilizer unit; acquisition of the nose pattern image by the
image acquisition unit;
storing of the acquired nose pattern image; generation of a nose pattern code
from the acquired
image; enrollment of the generated nose pattern code; and verification or
identification of the
subject animal by matching the newly obtained nose pattern code with
previously enrolled nose
codes.
Yet another technical solution of the present invention is to provide an
animal recognition
method comprising the following steps: acquisition of a nose pattern image
using the body
stabilizer unit and image acquisition unit; setting a region of interest (ROT)
in the acquired nose
pattern image, raw or processed; generation of a nose pattern code from the
ROT or standardized
ROT; enrollment of the newly generated nose pattern code; and verification or
identification by
determining the distance between the newly generated nose pattern code and
previously enrolled
nose codes.
6
CA 02925275 2016-03-23
Advantageous Effects
The present invention has an advantageous effect in the fast and accurate
recognition of
animals through the acquisition of nose patterns without making direct
physical contact with the
nose.
Another effect of the present invention is the acquisition of nose pattern
images fit for
recognition by utilizing a body stabilizer unit that minimizes obstructive
movements in a subject
animal that behaves uncooperatively out of fear or aggression toward the image
acquisition
equipment or the operator, and maintains the ideal frontal capturing angle on
the subject's nose.
Yet another effect of the present invention is the acquisition of good quality
nose pattern
images with the use of a body stabilizer unit designed to accommodate animals
of different sizes
and species.
Yet another effect of the present invention is the acquisition of high quality
nose pattern
images by utilizing indirect illumination of appropriate wavelength regions
applied through a
light conduit subunit, light diffuser subunit, and spacer onto the subject's
nose to prevent
unwanted light reflections that may come off the layer of moisture on the nose
surface of subject
animals.
Yet another effect of the present invention is the acquisition of high quality
nose pattern
images through the use of an image capture unit that is modifiable to
accommodate subject
animals of different species.
Yet another effect of the present invention is to enable non-professional
users to acquire
nose pattern images fit for recognition with ease using the image acquisition
unit.
Yet another effect of the present invention is to make possible the
identification of any
animal with a discernible nose pattern, regardless of species- or breed-
specific pattern types.
7
CA 02925275 2016-03-23
Yet another effect of the present invention is to make identification possible
regardless of
the presence of extraneous physiological or environmental phenomena, such as
moisture, hair, or
dust of the subject animal's nose surface.
Yet another effect of the present invention is to make identification possible
despite
reasonable variations in the image capturing angle.
Yet another effect of the present invention is to generate a universal nose
code irrelevant
species or breed for use in identification.
Yet another effect of the present invention is the use of the most appropriate
method of
verification or identification for the particular species or breed.
Yet another effect of the present invention is to increase the accuracy rate
of recognition for
each species of subject animals by comparing and matching previously stored
nose pattern
images to those newly obtained using the proper body stabilizer unit, image
acquisition unit and
image recognition unit.
Brief Description of the Drawings
FIG. 1 is a schematic diagram of the embodiment of the animal recognition
apparatus
described in the present invention.
FIG. 2 is a photograph of the nose pattern of a specific animal species (deer)
to demonstrate
an example of the subject of the present invention;
FIG. 3 is a photograph of the nose pattern of a specific animal species (dog)
to demonstrate
another example of the subject of the present invention.
FIG. 4 is a photograph of obstructive light reflections from the moisture
naturally present on
the surface of the nose of the subject animals.
FIG. 5 is a diagram illustrating the operation of the animal recognition
apparatus in the
present invention.
FIG. 6 is a presentation of the animal recognition apparatus in FIG. 5 shown
from different
angles.
8
CA 02925275 2016-03-23
FIG. 7 is a block diagram illustrating a configuration of the animal
recognition apparatus in
the present invention in which each of the parts, the body stabilizer unit,
the image acquisition
unit, and the image recognition unit are all separate;
FIG. 8 is a block diagram illustrating a configuration of the animal
recognition apparatus in
the present invention in which the body stabilizer unit and the image
acquisition unit are separate
from the image recognition unit;
FIG. 9 is a block diagram illustrating a configuration of the animal
recognition apparatus in
the present invention in which the image acquisition unit and the image
recognition unit are
separate from the body stabilizer unit;
FIG. 10 is a block diagram illustrating a configuration of the animal
recognition apparatus
in the present invention in which the body stabilizer unit, the image
acquisition unit, and the
image recognition unit are all connected.
FIG. 11 is a flowchart illustrating the method of operating the animal
recognition apparatus
in the present invention.
FIG. 12 is a block diagram schematically showing how to use the body
stabilizer unit;
FIG. 13 is a block diagram schematically showing how to use the posture
stabilizer unit;
FIG. 14 is a block diagram schematically showing how to use the position
adjuster unit.
FIG. 15 is a diagram illustrating the application of the upper and lower body
stabilizer units
on two specific species (cow and deer);
FIG. 16 is a diagram illustrating the application of the upper and lower body
stabilizer units
on two other specific species (dog and cat).
FIG. 17 is a diagram illustrating the configuration of the head stabilizer
unit.
FIG. 18 is a diagram illustrating the implementation of the appropriate
posture stabilizer
unit and position adjuster unit on a specific species (cow);
FIG. 19 is a diagram illustrating the implementation of the appropriate
posture stabilizer
unit and position adjuster unit on another species (deer);
FIG. 20 is a diagram illustrating the implementation of the appropriate
posture stabilizer
unit and position adjuster unit on two small species (dog and cat).
FIG. 21 is a flowchart illustrating a method of operating the body stabilizer
unit in the
present invention.
FIG. 22 is a block diagram illustrating the configuration of the image
acquisition unit.
9
CA 02925275 2016-03-23
FIG. 23 is a block diagram schematically illustrating the image capture unit
and the image
analysis unit within the image acquisition unit.
FIG. 24 is a block diagram illustrating the configuration of the image capture
unit that
moves the lens module and sensor according to the distance adjustment
principle of the distance
adjuster module.
FIG. 25 is a block diagram illustrating the configuration of the image capture
unit that
adjusts the distance between the lenses in the lens module according to the
distance adjustment
principle of the distance adjuster module.
FIG. 26 is a diagram illustrating the configuration of the front unit of the
image capture unit.
FIG. 27 is a diagram illustrating the configuration of the illumination unit.
FIG. 28 is a block diagram illustrating the method of obtaining nose pattern
images that are
usable by the image recognition unit through the image capture unit.
FIG. 29 is a diagram illustrating the method of adjusting the field of view
and focus by
moving the lens module or sensor in the image capture unit;
FIG. 30 is a diagram illustrating the method of adjusting the field of view
and focus by
moving the lenses within the lens module of the image capture unit.
FIG. 31 is a diagram illustrating how to manipulate the field of view adjuster
lens, the
length of the spacer, and the type of front unit to fit the noses of larger
subject animals (cow and
deer);
FIG. 32 is a diagram illustrating how to manipulate the field of view adjuster
lens, the
length of the spacer, and the type of front unit to fit the noses of medium-
sized subject animals
(dog);
FIG. 33 is a diagram illustrating how to manipulate the field of view adjuster
lens, the
length of the spacer, and the type of front unit to fit the noses of smaller
subject animals (cat or
very small dog).
FIG. 34 is a set of photographs showing the results of using the three
different types of the
front unit.
FIG. 35 is a pair of photographs comparing the results of using direct
illumination of a
conventional camera and the indirect illumination of the present invention to
acquire the nose
pattern image of the same individual.
CA 02925275 2016-03-23
FIG. 36 is a block diagram schematically describing the image analysis unit of
the present
invention.
FIG. 37 is a diagram illustrating the method of nose pattern image acquisition
during
capture mode.
FIG. 38 is a flowchart illustrating the method of nose pattern image
acquisition by the
image acquisition unit.
FIG. 39 is a block diagram schematically describing the image recognition unit
of the
present invention.
FIG. 40 is a flowchart illustrating the method of analyzing and recognizing
nose pattern
images.
FIG. 41 is a block diagram schematically describing the region of interest
fixing unit.
FIG. 42 is a diagram illustrating the method of finding the boundary of the
nostril.
FIG. 43 is a pair of diagrams illustrating the method of approximating the
boundary of the
nostrils with curves (circle and ellipse, respectively)
FIG. 44 is a diagram illustrating the method of obtaining the region on the
opposite side of
each nostril, that is located in the exterior of the approximated curves
(circle/ellipse).
FIG. 45 is a diagram illustrating the method of selecting the rectangular area
between the
approximation curves (circle/ellipse) as the region of interest.
FIG. 46 is a diagram illustrating the differences in the region of interest
resulting from using
circular and elliptical approximation curves on the same nose pattern image.
FIG. 47 is a diagram illustrating the process of generating a standardized
region of interest
from a previously fixed region of interest.
FIG. 48 is a simplified block diagram describing the nose pattern code
generation unit.
FIG. 49 is a block diagram illustrating the process of generating nose pattern
codes from the
region of interest.
FIG. 50 is a diagram illustrating how to divide the region of interest into
smaller cell blocks
of specified dimensions, from which frequency transform codes are generated.
FIG. 51 is a diagram illustrating the comparison of a theoretical calculation
area and the
actual area when generating the frequency transform code using Gabor
transform, Gabor cosine
transform, Gabor sine transform, etc.
FIG. 52 is a simplified block diagram describing the nose pattern code
matching unit.
11
CA 02925275 2016-03-23
FIG. 53 is a diagram illustrating the method of nose pattern code
identification through
simple matching.
FIG. 54 is a diagram illustrating a situation in which different regions of
interest have been
selected from the same nose pattern image of the same individual for matching.
FIG. 55 is a diagram illustrating the method of nose pattern code
identification through shift
matching.
FIG. 56 is a diagram illustrating a matching situation in which the regions of
interest of the
nose pattern code selected from the same individual have nonidentical vertical
and horizontal
proportions.
FIG. 57 is a diagram illustrating the method of nose pattern code
identification through
block-wise shift matching.
FIG. 58 is a diagram illustrating the process of nose pattern code
identification through shift
matching using Gabor sine transform.
FIG. 59 is a diagram illustrating the process of nose pattern code
identification through
block-wise shift matching using Gabor sine transform.
FIG. 60 is a diagram illustrating the method of nose pattern code
identification (one-to-
many matching).
Best Mode for Carrying Out the Invention
The following section describes the configuration and operation of the present
invention,
wherein the accompanying diagrams merely provide examples of one or more
possible
embodiments and do not limit the technical concept or its core components and
applications.
Therefore, for those with the skill and knowledge in the field of the present
invention should be
able to apply various changes and modifications to the presently described
embodiment of an
animal recognition apparatus without deviating from the core concept.
In explaining the components of the present invention, such terms as A, B,
(a), (b) and the
like can used. These are simply intended to distinguish one component from
another, and not a
reflection of the specific nature, order, or sequence of said components. When
a component is
described to be "connected to," "included in," or "configuring" another, they
may be directly
12
CA 02925275 2016-03-23
connected or coupled, but it should be understood that some other component
could also be
"connected to," "included in," or "configuring" each of the said components.
Also, for ease of understanding, when one component is shown in multiple
figures it will be
given a different reference numeral each time to correspond to the rest of the
diagram.
Furthermore, the present invention seeks to distinguish between the terms
Verification,
Identification, and Recognition. Specifically, Verification refers to one-to-
one (1:1) matching,
Identification or Searching refers to one-to-many (1:n) matching, and
Recognition emcompasses
both the Verification and Identification processes.
Several representative species -- the cow, deer, dog and cat -- with nose
patterns have been
selected here to illustrate the utilization of the present invention; and for
whenever the method or
apparatus is universally applicable, a single species is shown in the example
for sufficient
understanding. The implication is that the application of the present
invention pertains not only
to the specifically mentioned animals, but any and all species with distinct
nose patterns.
In the present invention, nose pattern relates to how the beads and grooves
form geometric
patterns on the nose surface, and it should be noted that the size and
intricacy of the patterning
can vary even within the same species.
The present invention, as outlined in FIG. 1, describes an animal recognition
method and
apparatus for animals with unique nose patterns (subject animals) through the
acquisition of
identifiable nose pattern images by utilizing a body stabilizer unit to
minimize movement and
resistance in the subjects; an image acquisition unit to capture said images;
and an image
recognition unit that generates processed nose pattern images via noise
reduction and image
quality reinforcement techniques, and from it, nose pattern codes for
enrollment and
identification.
The image acquisition unit 102 may include the image analysis unit to be
described later; or,
the image analysis unit may be included in the image recognition unit 103. In
other words,
13
CA 02925275 2016-03-23
various configurations and modifications are entirely possible to suit the
user's request or the
designer's purpose.
A significant number of species of animals are known to have unique nose
patterns. FIG. 2
and FIG. 3 show two examples of nose patterns, as taken using a model of the
image acquisition
unit. In the deer nose pattern in FIG. 2 the key features are the nostrils and
the beading and
grooving patterns, where beads are areas of raised nose skin and grooves are
the narrow valleys
surrounding each bead. FIG. 3 shows the nose pattern on a dog where, while the
specific size and
shapes differ, a similar beading and grooving phenomenon can be found.
In the case of cows, deer and other larger animals, the beads tend also to be
relatively larger,
while in smaller species like cats and dogs the beads tend to be
proportionately smaller. In fact,
even in the same species, the size of the nose area as generally dependent on
the body size can
affect the size of the beading patterns; and so it is necessary that the
recognition apparatus and
method take into consideration the patterning variations in different species
and breeds.
Next, it is imperative to address the physiological characteristics of the
nose of subject
animals. As shown in FIG. 4, a healthy nose maintains a layer of moisture on
the surface, which
aggressively reflects light in photos taken under natural settings. This also
adversely affects the
results of any method that relies on contact imprinting, such as with paper or
glass, as the
moisture can often cause blurring and image distortions. As for image
capturing, there is also
ample possibility that the moisture would reflect light or absorb the infrared
range. Thus, the
moisture presents an unavoidable problem that needs to be addressed and
solved.
The size and shape of the animal's face and body, as relating to the species
and breed, also
matter as the shape and length of the head and muzzle affect the capture and
body stabilizer units.
The temperament of the subject animal is a factor as well, as it can vary from
one individual
to another even in the same species or breed. While some subjects are
naturally tame and
cooperative, others are more timid or aggressive especially towards the
various (capturing,
stabilizing, or illumination) apparati and human operators, making the work
out in the field
14
CA 02925275 2016-03-23
difficult or unsafe, especially for a non-professional user. Therefore, an
effective method and
apparatus must not aggravate the subject animals.
The technical configuration of the animal recognition apparatus is as follows:
body
stabilizer unit, image acquisition unit, and image recognition unit. The body
stabilizer unit refers
to the set of devices that prevent disruptive movements from the subject
animal; the image
acquisition unit refers to the software and hardware described in the present
invention necessary
to acquire the nose pattern images of a variety of animals; and the image
recognition unit refers
to the software and hardware needed for nose pattern image recognition.
FIGS. 5 and 6 illustrate a practical application and operation of the overall
animal
recognition apparatus in the present invention, whereas the block diagrams in
FIGS. 7, 8, and 9
show the connective combinations among the body stabilizer, image acquisition,
and image
recognition units. As previously mentioned, the animal recognition apparatus
may be configured
with a certain level of flexibility depending on the given animal or setting,
where all three
component units could be connected, or just two, or all three are set up
separately.
The flowchart in FIG. 11 summarizes the method of animal recognition in the
present
invention, starting with S1101 selecting and S1102 fitting the animal into the
appropriate body
stabilizer unit; S1103 fixing the nose of the subject onto the image
acquisition unit and S1104
acquiring the nose pattern image; S1105 at the image recognition unit,
generating a nose pattern
code using the raw or processed nose pattern image, and S1106 enrolling and
identifying the
individual using the nose pattern code. However, this is not to say that this
sequence of events
cannot be modified; those knowledgeable in the field of the present invention
may choose to
change the order of the steps or run more than one in parallel without
departing from the core
concept.
The purpose of the body stabilizer unit is to temporarily control the movement
or resistance
of the subject animal in reaction to the illumination or the operator, such as
head turning or
aggressive behavior, during the nose pattern image acquisition process to
yield the best quality
image. This is a safety measure against the animal accidentally harming itself
or the human
CA 02925275 2016-03-23
operators, which would incur added difficulties as well as cost. Moreover, an
ideal nose pattern
image is one taken from head-on and this is difficult to obtain with a highly
uncooperative
subject animal without the help of a body stabilizer.
Thus, the four primary functions of the body stabilizer unit are as follows:
minimize the
motion of the subject animal during the image acquisition process, act as a
safety measure to
protect the operator and apparatus, protect the subject animal from self-harm,
and hold the nose
in place for the best angle of image capture. As shown in FIG. 12, the body
stabilizer unit
comprises the posture stabilizer unit 1201 , and also the position adjuster
unit 1202 to
accommodate the subject's stance width. FIG. 13 shows a more detailed
breakdown of the
posture stabilizer unit.
Although it is easy to assume that the only body part that needs to be
properly restrained to
obtain a good nose image is the head, the subject animal can and often will
resist with the whole
body and thereby cause blurry images. This problem may be mitigated by
stabilizing the neck
and shoulder area (upper body), as well as the back, front, and hind legs
(lower body). Vets
commonly forgo the usage of anesthesia during procedures whenever possible by
applying
pressure on the nape or shoulder of the patient animals; the body stabilizer
is meant to simulate
this method by allowing the subject animal to rest its head on the chin
support while holding it in
position with the head stabilizer unit 1302 and applying appropriate pressure
using the upper
body stabilizer unit 1301. Further movement in the lower body, especially in
the case of larger
animals whose powerful legs may pose a danger to the operators and equipment,
may
additionally be held in check by the lower body stabilizer unit 1303.
The configuration of the position adjuster unit is modifiable in accordance
with the posture
stabilizer unit settings, as the operator sees fit for the subject animal.
Possible additions are the
height adjuster unit 1404 to the upper body stabilizer unit 1401; stance width
adjuster unit 1406
to the lower body stabilizer unit 1403; and the horizontal balance adjuster
unit 1405 and the
height adjuster unit 1407 to the head stabilizer unit 1402.
16
CA 02925275 2016-03-23
FIGS. 15 and 16 each show an example of the posture stabilizer unit with upper
and lower
body stabilizer units as appropriate for larger animals as cows and deer, and
for smaller animals
as dogs and cats, respectively. Depending on the species, the upper and lower
body stabilizers
may be set up in various combinations -- each independently, in conjunction,
or connected at
certain parts.
In both FIGS. 15 and 16, the upper body stabilizer unit comprises the upper
body stabilizing
brace subunit 1501, 1601 and the upper body stabilizing brace lock subunit
1502, 1602; the
upper body stabilizing brace pressure adjuster subunit 1503, 1603 is optional.
The upper body
stabilizing brace subunit 1501, 1601 may be made into a cover type, with
durable synthetic
fabrics, or with length-adjustable belts or cables. The upper body stabilizing
brace lock subunit
1502, 1602 prevents the brace subunit from coming undone during the procedure,
and may be
manual or electronic. The upper body stabilizing brace pressure adjuster
subunit 1503, 1603
allows the upper body stabilizing brace subunit to apply pressure on the
subject animal by, for
example, inflating the brace with some gas or liquid with the use of a
pressure injector paired
with a pressure monitor subunit.
Likewise, in both FIGS. 15 and 16, the lower body stabilizer unit comprises
the lower body
stabilizing brace subunit 1504, 1604 and the lower body stabilizing brace lock
subunit 1505,
1605; the lower body stabilizing brace pressure adjuster subunit 1506, 1606,
as well as the lower
body stabilizer supporting subunit 1507, 1607 are optional. The lower body
stabilizing brace
subunit 1504, 1604 may be made into a cover type, with durable synthetic
fabrics, or with
length-adjustable belts or cables. The lower body stabilizing brace lock
subunit 1505, 1605
prevents the brace subunit from coming undone during the procedure, and may be
manual or
electronic. The lower body stabilizing brace pressure adjuster subunit 1506,
1606 allows the
lower body stabilizing brace subunit to apply pressure on the subject animal
by, for example,
inflating the brace with some gas or liquid with the use of a pressure
injector pair with a pressure
monitor subunit. The lower body stabilizer supporting subunit 1507, 1607
fastens the lower body
stabilizer unit to the ground or at a certain distance from the equipment, and
may be made up the
lower body supporting subunit and the lower body supporter connector subunit.
The lower body
supporting subunit and the lower body supporter connector subunit may take
many forms to suit
17
CA 02925275 2016-03-23
the subject animal, and may be made of steel or other metals, as well as
durable synthetic fibers,
rubber, or fabric.
The head stabilizer unit in FIG. 17 comprises the chin support 1701, and the
stabilizing
muzzle subunit 1702 that holds the subject's nose in the correct position. The
chin support 1701
may be made of various materials as wood, plastic, rubber, or metal, and
should withstand the
weight of the subject's head while providing a comfortable headrest and room
for additional
supporting attachments. The stabilizing muzzle subunit 1702 will be used when
the head
movement cannot be controlled with the chin support alone, and may be made
into a cover type
with durable synthetic fabrics, or with length-adjustable belts or cables, to
span the muzzle area.
The stabilizing muzzle lock subunit 1703 prevents the brace subunit from
coming undone during
the procedure, and may be manual or electronic. The stabilizing muzzle
pressure adjuster subunit
1704 allows the stabilizing muzzle subunit to apply pressure on the subject
animal's muzzle by,
for example, inflating the brace with some gas or liquid with the use of a
pressure injector paired
with a pressure monitor subunit. The head stabilizer unit may also have a head
stabilizer support
1705 that fastens the head stabilizer unit to the ground or at a certain
distance from the
equipment while supporting the weight of the subject animal's head, and may be
made into
various shapes using durable materials as wood, stone, or metal.
The position adjuster unit adjusts the settings of the posture stabilizer unit
as per each
animal's physical characteristics in order to produce the most comfortable
position for the
subject, and comprises the height adjuster unit, horizontal balance adjuster
unit, and stance width
adjuster unit. The examples in FIGS. 18 and 19 show the subject animal (cow
and deer,
respectively) fitted into the upper and lower body stabilizer units where the
height adjuster units
1801, 1804, 1901, 1904 are set to accommodate the subject's height. The height
adjuster 1801,
1901 connects the upper and lower body stabilizer units and may comprise the
connector subunit
1802, 1902 made with belts or cables, and the length adjuster subunit 1803,
1903 that adjusts the
length of the connector subunit. The height adjuster unit 1804, 1904 for the
head stabilizer unit
may also comprise the connector subunit 1805, 1905 connecting the chin support
to the ground
and the chin support height adjuster unit 1806, 1906. The horizontal balance
adjuster unit 1807,
1907, placed inside or outside the chin support and comprising a horizontal
balance sensor with a
18
CA 02925275 2016-03-23
display monitor, positions the chin support under the subject animal's head to
directly face the
image acquisition unit. The horizontal balance sensor may comprise gravity,
gyro, or pressure
sensors. The stance width adjuster unit 1808, 1908 may be used when the lower
body is fastened
in the lower body stabilizer unit, and may comprise connector subunit 1809,
1909 of belts or
cables and a length adjuster subunit 1810, 1910 that connects both sides of
the lower body.
As shown in FIG. 20, for certain species of animals like cats or dogs, it may
be appropriate
to combine the upper and lower body stabilizer units with a height adjuster
unit 2001, placed in
front or on top of the head stabilizer unit, that positions the chin support
to fit the height of the
subject animal and reach its head. The height adjuster unit 2001 that adjusts
the height of the
head stabilizer unit may comprise a connector subunit 2002 that connects the
chin support to the
ground and a height adjuster subunit 2003 that adjusts the height of the
connector subunit. The
horizontal balance adjuster unit 2004, placed inside or outside the chin
support and comprising a
horizontal balance sensor with a display monitor, positions the chin support
under the subject
animal's head to directly face the image acquisition unit. The horizontal
balance sensor may
comprise gravity, gyro, or pressure sensors. The stance width adjuster unit
2005 may be used
when the lower body is fastened in the lower body stabilizer unit, and may
comprise a connector
subunit 2006 of belts or cables and a length adjuster subunit 2007 that
connects both sides of the
lower body.
The sequence of operation for the body stabilizer unit is as follows: S2101
select the
appropriate body stabilizer unit for the subject animal by taking into
consideration the overall
size, leg length, feet size, head size, and the relative location of the nose;
S2102 fit the subject
animal into the upper body stabilizer unit; S2103 fasten the upper body by
utilizing the upper
body stabilizing brace subunit and upper body stabilizing brace pressure
adjuster subunit to fit
the shoulder width; S2104 fit the subject animal into the lower body
stabilizer; S2105 fasten the
lower body by utilizing the lower body stabilizing brace subunit and lower
body stabilizing brace
pressure adjuster subunit to fit the ankles or legs; S2106 set the stance
width adjuster, and also
the height adjuster to fit the subject's height if necessary to connect the
upper and lower body
stabilizer units; S2107 fasten the head by utilizing the head stabilizer unit,
making sure to set the
height adjuster unit to the correct height and the horizontal balance adjuster
unit to have the nose
19
CA 02925275 2016-03-23
facing the image acquisition unit head-on. This sequence of events may be
modified, by those
knowledgeable in the field of the present invention, without departing from
the core concept.
The purpose of the image acquisition unit is the capture and acquisition of
nose pattern
images. This can seem conceptually innocuous but the execution of it is
nothing but, due to the
morphological diversity of the nose and nose patterns, as well as the
physiological nature of the
rhinarium that yields unwanted light reflections. Thus, the six primary
functions of the image
acquisition unit are as follows: acquire good quality nose images usable by
the image recognition
unit without relying on the traditional methods that mandate direct contact;
acquire good quality
nose images from a wide variety of species; not be affected by a subject
animal's particular size,
shape, or physiology; employ a special kind of illumination to avoid issues
with light reflections
from the wet nose; and enable non-professional users to achieve the above five
with ease.
As shown in FIG. 22, the image acquisition unit comprises the image capture
unit 2201 that
photographically captures nose pattern images, and also possibly the image
analysis unit 2202
that analyzes the captured images and processes certain signals and
information. The image
capture unit comprises the capture unit 2301, the front unit 2302 that adjusts
the field of view
(FOV) and capture distance for each subject while blocking out the ambient
light for a more
controlled environment, and additionally the illumination unit 2303 that
provides indirect
illumination to overcome the light reflection issue (FIG. 23).
The image capture unit, as illustrated in FIG. 24, may comprise the lens
module 2401 with
two or more lenses; an image sensor 2402 (CMOS or CCD); and the distance
adjuster module
2403 that controls the FOV and focus by moving the lens module and sensor,
thereby
manipulating the distances between the lenses and between the lens module and
the sensor
within the front unit 2404. The distance adjuster module 2401 moves the lens
module or a
plurality of lenses, and comprises a small motor and rack gear that the
converts motor's circular
motion to linear motion. Also, a guide rail that allows the lens module 2401
and sensor 2402, in
linear periodic motion by the rack gear, to move between predetermined
positions may also be
installed. Alternatively, the image capture unit may also comprise the lens
module 2501 and
CA 02925275 2016-03-23
sensor 2502 in fixed positions, with the distance adjuster module 2503 only
controlling the
distances between the lenses within the lens module (FIG. 25).
The front unit may comprise a front cover 2601 that surrounds and/or comes in
contact with
the skin around the nose when the nose enters the front unit; a FOV adjuster
lens 2603; a spacer
2602 that adjusts the distance between the subject's nose and the FOV adjuster
lens 2603. The
front cover 2601 and spacer 2602 may come in variable shapes or sizes to
accommodate
different species or breeds. The front cover 2601 should be of a color that is
best suited for
blocking out ambient light, most likely black or other dark hues, and made of
materials that do
not agitate the subject animals, such as synthetic fibers, rubber, textile, or
plastic. The front cover
also may be imbued with a calming scent for the subject animals, and made to
be detachable for
easy substitution when dealing with subjects of different physical
requirements during the same
session.
The standard FOV adjuster lens 2603 is modeled after the nose size of a
typical (medium
sized) dog; a reducing lens is used instead for larger noses, and a magnifying
lens for smaller
noses. The standard lens refers to a single or a set of lenses that allows the
framing of a typical
dog nose, and the reducing and magnifying lenses are made in relation to the
standard.
The spacer 2602 consists of the exterior that the blocks the light coming from
the outside,
and the interior that surrounds the nose of the subject animal, and possibly
also houses an
illumination unit. The length of the spacer, which determines the distance
between the FOV
adjuster lens and the subject's nose, may be optimized using field trial
results. It also may be
efficient to have pre-designed, detachable front units with spacers and FOV
adjuster lenses set to
fit particular species or nose sizes based on experimental results.
The illumination unit in FIG. 27 seeks to eliminate the issues that arise from
the reflection
and absorption of light by the moisture on the nose surface by incorporating a
light source 2701
of a specific wavelength region (that poses no threat to the health of the
subject animal) in
indirect illumination, wherein the light travels through a light conduit
subunit 2703 and light
diffuser subunit 2702. The light source 2701, light diffuser subunit 2702, and
light conduit
21
CA 02925275 2016-03-23
subunit 2703 may vary to suit different species of subject animals. The light
source 2701 should
have adjustable luminosity, avoid the infrared region that can be absorbed by
the moisture and
the UV region that can cause tissue damage, and be optimized to suit the
particular
characteristics of a species. Any type of light source consistent with the
above description would
suffice.
The light diffuser subunit 2702 partially absorbs and reflects light from the
light source
through the diffuser surface to indirectly illuminate the whole nose surface
inserted into the front
unit. The amount of light that eventually passes through the diffuser may be
controlled with the
type of material used, such as Hanji(traditional Korean paper handmade from
mulberry trees),
translucent tracing paper, or a special type of glass, and similar material
may also be used to line
the interior of the light conduit subunit 2703.
In general, to obtain a high quality image, correctly setting the depth of
field is important;
for that purpose it is typical to adjust the aperture, FOV, focus, and the
distance to the subject.
The image capture unit 2801 employs a variety of ways to obtain good quality
images. Within
the capture unit 2802, the FOV and focus are controlled by the distance
adjuster module 2805
either by moving the lens module 2811A or sensor 2812, or by moving the
plurality of lenses
2813 within the lens module while the lens module 2811B and sensor 2812 stay
fixed. The front
unit 2803 adjusts the FOV with the FOV adjuster lens 2806, and the focus by
changing distance
between the FOV lens 2806 and lens module 2815 via variable spacer 2807
length. The
illumination unit 2804 employs a light source 2808 of a wavelength region
optimal for nose
images and the light conduit subunit 2809 and light diffuser subunit 2810 for
indirect
illumination, which is essential to producing good quality images without
obstructive reflections
(FIG. 28).
There are two different ways the capture unit adjusts the FOV and focus. The
first method,
as shown in FIG. 29, involves moving the lens module 2901 or the sensor 2902,
independently or
concurrently, along a linear axis using the distance adjuster module. The
change in the position
of the lens module 2901 or sensor 2902 changes the distance (a) between the
two, and the
distance (b) between the lens module 2901 and the FOV adjuster lens within the
front unit 2903,
22
CA 02925275 2016-03-23
thereby changing the FOV and focus. The length of the spacer 2903 is preset
for the particular
subject animal, and the distances (dl, d2) between the lenses 2904 in the lens
module are also
fixed. The values of a and b could also be set in advance for specific species
so that non-
professional users could easily carry out the capture process.
The second method, as shown in FIG. 30, involves moving the lenses 3004 within
the lens
module along a linear axis, thereby changing the distances between the lenses
(dl, d2), to change
the FOV and focus. Meanwhile, the length (c) of the front unit 3003 is set in
advance for the
appropriate species and therefore a fixed value; and the distance (a) between
the lens module
3001 and the sensor 3002, and the distance (b) between the lens module and the
FOV adjuster
lens in the front unit 3003 are also fixed. Moreover, the distance adjuster
module may be
configured to move with the lenses 3004 within the lens module so that only dl
and d2 values
can be manipulated. Again, the values of a and b could also be set in advance
for specific species
so that non-professional users could easily carry out the capture process.
The front unit uses the FOV adjuster lens and the length of the spacer to
manipulate the
FOV and focus. FIGS. 31, 32, and 33 illustrate the how different combinations
of the FOV
adjuster lens and spacer length may be used to accommodate subject animals of
different sizes.
As mentioned previously, the standard FOV adjuster lens is modeled after the
nose size of a
typical (medium sized) dog; a reducing lens is used instead for larger noses,
and a magnifying
lens for smaller noses. The standard lens refers to a single or a set of
lenses that allows the
framing of a typical dog nose, and the reducing and magnifying lenses are made
in relation to the
standard. Also, the length of the spacer may be changed, depending on the nose
size of the
subject animal, to adjust the focus (distance).
For subject animals with a large nose surface area, like the cow in FIG. 31,
using a standard
lens calibrated for the size of a typical dog would not yield a large enough
FOV to capture the
whole nose and could negatively affect the recognition accuracy. Therefore,
the FOV adjuster
lens 3101 should be a reducing lens, and the spacer 3102 should be set in
advance to a length
appropriate to get the right focus on the subject animal. Also, since the
length (c) of the spacer
3102 can change, the distance (a) between the lens module 3103 and the sensor
3104 and the
23
CA 02925275 2016-03-23
distance (b) between the lens module and the FOV adjuster lens in the front
unit 3101 may also
change.
For subject animals with a medium-sized nose surface area, like the dog in
FIG. 32, using a
reducing lens would widen the FOV too much and yield images that do not show
the level of
detail in the nose pattern image necessary for accurate identification.
Conversely, using a
magnifying lens would reduce the FOV too much and not capture the whole nose,
diminishing
the amount of nose pattern data obtained. Therefore, a standard FOV adjuster
lens 3201 should
be used for medium-sized noses, paired with an appropriately lengthed spacer
3202. Also, since
the length (c") of the spacer can change, the distance (a") between the lens
module 3203 and the
sensor 3204 and the distance (b") between the lens module and the FOV adjuster
lens in the
front unit 3201 may also change.
For subject animals with a small nose surface area, like cats or the small dog
in FIG. 33,
using the standard lens would widen the FOV too much and yield images that do
not show the
level of detail in the nose pattern image necessary for accurate
identification. Therefore, the FOV
adjuster lens 3301 should be a magnifying lens, and the spacer 3302 should be
set in advance to
a length appropriate to get the right focus on the subject animal. Also, since
the length (c") of
the spacer can change, the distance (a") between the lens module 3303 and the
sensor 3304 and
the distance (b") between the lens module and the FOV adjuster lens in the
front unit 3301 may
also change.
FIG. 34 shows the results of using the different front unit settings on a dog,
demonstrating
the importance of choosing the right one; the first image was taken with the
magnifying setting,
the second with the standard, and the third with the reducing.
The next step to improving the image quality is in the illumination. FIG. 35
shows a side-
by-side comparison of the same dog nose image captured using the conventional
direct
illumination (camera flash) and the indirect illumination of the illumination
unit.
24
CA 02925275 2016-03-23
The illumination unit controls the light reflections -- which appear as white
flecks on the
left image, and are highly obstructive to accurate identification -- from the
moisture on the nose
surface by achieving indirect illumination through the use of a special light
source, light conduit
subunit, and light diffuser subunit. The light source should avoid the
infrared region that can be
absorbed by the moisture and the UV region that can cause tissue damage, and
be optimized to
suit the particular characteristics of a species.
Also, FIG. 35 demonstrates that using the conventional camera flash does not
help to
contrast the nostril from the nose surface area, while indirect illumination
results in clear
boundary distinctions. This affects the ease with which the nostril boundary
can be established,
and thus indirectly illuminated images will generally increase the recognition
accuracy.
The image analysis unit analyzes the nose pattern images acquired by the image
capture unit,
manages various kinds of information and signals from the process, and may
also be attached not
only to the image acquisition unit but also the image recognition unit. As
illustrated in the block
diagram in FIG. 36, the image analysis unit may comprise the main processor
unit 3601, the
buffer 3602, the database (DB) 3603, and the communication unit 3604. Also a
display unit 3605
may be added so that the operator may see the images captured by the image
capture unit in real
time, and select and acquire good quality images.
The main processor unit selects nose pattern images that are of sufficient
quality to be used
in the image recognition unit, out of all the images captured by the image
capture unit. When
multiple nose pattern images are obtained by the image capture unit, each
image is given
individual scores on specific variables, and images that pass the threshold
set by the image
analysis unit are selected. If none out of a particular batch meet the
threshold, then that whole
group is discarded and a request for a new batch is sent to the image capture
unit. During this
selection process, the images are evaluated on such criteria as, the amount of
light reflection,
sharpness, contrast ratio, ROI for capture, noise level, etc; and only those
images that pass the
threshold for each variable are accepted. When more than one image out of a
single batch pass
the threshold the one with the highest total score (sum of individual scores)
is selected, and this
CA 02925275 2016-03-23
process may take place simultaneously as the image acquisition in the image
analysis unit or the
image recognition unit.
There are two types of variables: those that are not related to species-
specific characteristics
(Al-A3) and those that are (A4-Al2). The former includes sharpness Al,
contrast A2, and noise
level A3; the latter includes ROT for capture A4, presence of light reflection
AS, nostril location
A6, sharpness of nostril image A7, contrast level of nostril image A8, noise
level of nostril image
A9, sharpness of the border between the nostril and ROI A10, contrast level at
the border
between the nostril and ROI All, and noise level at the border between the
nostril and ROI Al2.
Variables may be appropriately added to or subtracted from the above list
depending on a subject
animal species' particular characteristics(Table 1).
Table 1. Nose pattern image evaluation variables
variables Score
(A 1 ) sharpness al
(A2) contrast a2
(A3) noise level a3
(A4) ROT for capture a4
(A5) presence of light reflection a5
(A6) nostril location a6
(A7) sharpness of nostril image a7
(A8) contrast level of nostril image a8
(A9) noise level of nostril image a9
(A 10) sharpness of the border between the nostril and ROI al0
(All) contrast level at the border between the nostril and ROT all
(Al2) noise level at the border between the nostril and ROT a 1 2
When calculating the total score to select the best (highest scoring) image
out of a batch that
yields more than one image that pass the threshold, if the numerical value of
sharpness is al, and
the weight factor of this is wl; contrast is a2, and w2; noise level is a3,
and w3; ROT for capture
is a4, and w4; presence of light reflection is a5, and w5; nostril location is
a6, and w6; sharpness
26
CA 02925275 2016-03-23
of nostril image is a7, and w7; contrast level of nostril image is a8, and w8;
noise level of nostril
image is a9, and w9; sharpness of the border between the nostril and ROI is
al0, and w10;
contrast level of the border between the nostril and ROT is all, and w11; and
noise level at the
border between the nostril and ROT is a12, and w12; then the total score is
the sum of the
products of al and wl, a2 and w2, a3 and w3, a4 and w4, a5 and w5, a6 and w6,
a7 and w7, a8
and w8, a9 and w9, al0 and w10, all and wl 1, and a12 and w12, is expressed
using the
following formula:
Total Score = wl*al + w2* a2 + w3* a3 + w4* a4 + w5* a5 + w6* a6+ w7* a7+ w8*
a8
+w9* a9+ w10* al0 + wll* all +w12* al2 -------------- (Equation 1)
The above total score is the weighted sum of the individual scores, and
therefore the degree
of importance of a particular variable may be reflected by adjusting the
weight value.
There are two ways the image acquisition unit can capture a batch of nose
pattern images.
When the image capture unit is not on standby in sleep mode, it is in
automatic or manual
capture mode. Automatic capture mode receives the threshold values for the
species or breed
from the DB, and compares them to the individual scores of the captured images
at the main
processor unit. On the other hand, in manual mode the user operates the image
acquisition unit,
visually makes an estimated evaluation of the individual scores of each
variable and makes the
capturing decision if these scores are deemed satisfactory. The sleep mode is
a standby mode
before entering the capture (recording) mode, and the capture mode is for the
final nose pattern
image acquisition. The image acquisition unit may transition from sleep mode
to capture mode
when the user presses a designated button on the display unit.
FIG. 37 illustrates the acquisition of a batch of nose pattern images in
capture mode. Once
in capture mode, the lens module or sensor in the capture unit moves into
position according to
the preset value. The transition from capture mode to sleep mode occurs when
the best
(threshold-passing) image is successfully selected by the main processor unit
from among the
batch saved to the buffer. If the time at the start of the recording is
T_start and the end of the
recording is T_end, then an n number of images are acquired during that time
at a constant rate
27
CA 02925275 2016-03-23
per second. The per second frame rate will vary depending on the resolution,
and may also vary
depending on the hardware configuration and the type of camera. The main
processor unit may
also alert the operator of the end of the capture process through a
notification on the display unit.
The minimum hardware components of the main processor unit are the CPU, RAM,
and
nonvolatile memory (ROM, flash memory, etc). The CPU performs all of the
operations carried
out by the image analysis unit. On the nonvolatile memory is mounted the
resident program
where the threshold values are stored, the individual scores of nose pattern
images are evaluated,
and the algorithm that enables the saving of the selected images and all
related information to the
buffer is stored. Also, when the computation for the selection of the best
image is too
complicated or when dealing with a very large number of images carried over
from the image
acquisition unit, nonvolatile memory may not be efficient for speedy
processing, in which case
RAM may be a useful addition to the main processor unit.
The buffer stores a variety of information that arise while the main processor
unit is in the
process of selecting threshold-passing images, and may consist of a flash
memory or a DB. Since
the DB on the buffer can be changed any time by the user, the DB of the buffer
generated by the
image analysis unit should preferably be stored in the flash memory. The
parameter DB stores
the threshold values and individual scores selected by the main processor unit
in the image
acquisition unit.
The communication unit relays information between the image capture unit and
image
analysis unit. The communication unit is tasked with the output of signals for
positioning
commands during capture and alerting the user of mode changes, and is thus
basically equipped
with a signal transmitter for outputting instruction signals. The signal
transmitter may comprise
one or more of the following: audible signal generator (for voice or other
sounds), visual signal
generator (for LED or flash), and vibration generator. Also, a display unit
possibly comprising a
mirror or LCD may be supplemented to enable a quick and easy review of the
images obtained
by the image capture unit.
28
CA 02925275 2016-03-23
The processes of the image analysis unit described thus far do not require a
separate OS, but
an appropriate one may be installed if there arises a need for the management
of memory or
other resources, time synchronization, or CPU time scheduling. Also, all of
the above mentioned
hardware can be mounted on a single chip for a one-chip solution (one-chip
microprocessor).
An example of the image acquisition process is illustrated in FIG. 38; the
order of events
need not be limited as follows. S3801 the operator selects the species on the
display unit to start
the automatic mode, or chooses the manual mode; S3802 in automatic mode, once
the species
selection is made, pressing the capture button starts the acquisition of the
batch of nose pattern
images at n frames per second while the lens module is shifted about within
the preset range of
positions (adjusting values of a and b). In the automatic mode, the image
capture unit,
illumination unit and front unit are automatically adjusted to accommodate the
subject animal
based on the values stored in the parameter DB for FOV, focus, luminosity,
etc. However, in
manual mode, the operator visually evaluates the features and variables of the
nose pattern
images through the display unit, and selects the best image. S3803 The nose
pattern images
acquired from the sensor of the image capture unit are stored in the buffer,
upon which the main
processor unit calculates the individual scores and compares to the threshold
values in the
reference DB. S3804 Once the best image that passes all the thresholds is
selected, it is stored in
the buffer.
As previously mentioned, the image recognition unit generates processed nose
pattern
images and nose pattern codes for enrollment and identification. In order to
successfully deal
with a wide range of variances, the image recognition unit should be capable
of the the following:
identify any individual animal with viable nose patterns, regardless of
idiosyncratic
characteristics; identify regardless of the extraneous physiological phenomena
(such as moisture
or hair, etc); compensate for certain distortions that occur in images
captured from different
angles and perform accurate identification; create universal nose pattern
codes for identification
for any species or breed with viable nose patterns; and employ the best method
of identification
when performing matching within a known species or breed.
29
CA 02925275 2016-03-23
As shown in FIG. 39, the image recognition unit may comprise the region of
interest (ROT)
fixing unit 3901, the nose pattern code generation unit 3902 that generates
the nose pattern code
from the fixed ROT, the nose pattern code matching unit 3903, and the nose
pattern code
database (DB) 3904 where the generated nose pattern codes are stored during
the enrollment and
identification stage. Possible additions are the image processing unit 3905,
which processes the
nose pattern image, if necessary, before setting the ROT; and the standardized
ROT fixing unit
3906, which standardizes the ROT before setting the ROT and generating the
nose pattern code.
Moreover, the aforementioned image analysis unit of the image acquisition unit
may be
configured into the image recognition unit.
FIG. 40 illustrates the method by which an animal nose pattern image is
analyzed to be used
for identification, the order of which may be modified to better suit the
equipment or
circumstances: S4001 acquisition of the subject animal's nose pattern image by
utilizing the
body stabilizer unit and image acquisition unit; S4003 setting the ROT on the
(processed) nose
pattern image; S4005 generating a nose pattern code from the fixed ROT; S4006
enrolling the
generated nose pattern code; S4007 comparing the stored nose pattern code from
the enrollment
to the newly generated nose pattern code in one-to-one matching for
verification; and S4008
running one-to-many matching for identification. Images acquired from S4001
that have been
processed are called processed nose pattern images, and an additional step
S4002 for storing
them may be included. Also, the step S4004 that generates a standardized ROT
from the ROT
selected in S4003 may also need to occur.
The image processing unit processes the acquired nose pattern images in order
to increase
the identification rate, and stores the resulting image. Raw acquired images
may present different
levels of noise and blurring, and may require contrast adjustments to
normalize the distribution
of pixel values. The present invention uses the histogram equalization
technique to normalize the
distribution of pixel values of images. In order to adjust the distribution of
pixel values, a
distribution function is fixed and histogram equalization is applied to each
nose pattern image to
have the same fixed distribution function. Image filtering techniques may also
be applied to take
care of the noise and blurring issues, with Gaussian or median filters for
noise level adjustment,
and with a variety of low-pass filters in the frequency domain.
CA 02925275 2016-03-23
Moreover, sharpening techniques using derivatives can be used to accentuate
the embossed
nose patterns, and de-convolution techniques can be used to restore damaged
images. In the
present invention, except when necessary to distinguish between the raw nose
pattern images and
the processed nose pattern images, both will be commonly referred to as nose
pattern images for
the sake of simplicity.
FIG. 41 is a schematic diagram briefly illustrating the ROT fixing unit as one
embodiment of
the present invention. The ROT fixing unit, as shown in FIG. 41, may comprise
the segmentation
unit 4101, the curve approximation unit 4102, and the ROT dividing unit 4103.
The segmentation
unit sets the boundaries of the nostrils, which become the basis for setting
the ROI in the nose
pattern image.
FIG. 42 is a schematic diagram illustrating how to find the nostril boundary
as one
embodiment of the present invention. FIG. 42 illustrates the nostril boundary
setting process,
where the nostrils appear as a shade due to the indirect illumination. The
boundary of this shade
is the basis for the nostril boundary, which may take the form of a circular
or elliptical arc, etc. In
order to extract the boundary points, starting with a point(s) within the
shade as the center
point(s), the boundary points are located based on the change in brightness
along the ray from
the fixed center points. Points along the rays extending in various directions
that display a sharp
change in brightness are marked as candidate points, and the correct boundary
points are found
among those candidate points based on the statistical analysis of nostril
shape and location.
Using the above statistical analysis, not all of the boundary points of the
nostril in various
directions may be extracted resulting in that only parts of boundary points
are extracted.
Sometimes, even with indirect illumination certain areas that are not part of
the nostrils may
appear to be inside of a similar shade, and therefore it is helpful to use
multiple center points and
statistical analysis utilizing the shape and location information of nostrils
to prevent finding
incorrect boundary points.
31
CA 02925275 2016-03-23
The curve approximation unit approximates the boundary curves of nostril
boundaries using
the boundary points found in the segmentation unit. In this approximation, the
final
approximation curve is the best curve fitting the boundary points found by
various regression
analyses, and it is usually a circular arc or elliptical arc.
FIG. 43 is a diagram illustrating how to approximate the nostril boundaries
with circles or
ellipses as one embodiment of the present invention. Although, as shown in
FIG. 43, the left and
right nostril boundaries can be regarded as symmetric curves when they are
seen from the front
of the nose, the two approximation curves can be asymmetric ellipses if the
nose pattern image is
taken from askew.
Also, since the curve approximation unit separately approximates the left and
right nostril
boundaries, the two approximation curves can have different shapes resulting
in that one curve is
a circle, and the other an ellipse. It is also possible that the two
approximation curves are
different in size although they are all either circles or ellipses.
The ROT dividing unit extracts a quadrilateral region of a nose pattern image
between the
two approximation curves obtained from the approximation unit. This process
consists of two
steps: a) the region between two approximation curves is identified and b) a
quadrilateral region
contained in the identified region is extracted.
(A) The first step where the region between two approximation curves is
identified:
FIG. 44 is a schematic diagram illustrating how to identify the region between
the two
approximation curves (circles or ellipses) as one embodiment of the present
invention. As shown
in FIG. 44, two points which are on the intersections between each
approximation curve and the
line segment connecting two centers of the approximation curves are located,
and the two
tangent lines which tangent at each located point to the approximation curve
(the left tangent line
is denoted by T_L, and the right tangent line by T R) are found. These tangent
lines may be
perpendicular to the line segment connecting the two centers when the two
approximation curves
are symmetrical, and may not be perpendicular when they are not symmetrical.
32
CA 02925275 2016-03-23
The two connecting lines are then found: one line connecting two upper vertex
points of the
approximation curves and the other line connecting two lower vertex points
(the upper line is
denoted by T_U, and the lower line denoted by T_D ). In this step, the two
connecting lines are
tangent lines which tangent to the both of the approximation curves when they
are both circles
and the two lines connect two upper vertex points or two lower vertex points
when they are both
ellipses.
(B) The second step where the quadrilateral region between two approximation
curves is
extracted as the ROI:
FIG. 45 is a schematic diagram illustrating how to extract the quadrilateral
region between
the two approximation curves as one embodiment of the present invention. As
shown in FIG. 45,
the ROT is the quadrilateral region encompassed by four lines obtained in Step
A. The shape and
the size of the ROT may be varied depending on the relative position of the
nose to the position
of the image acquisition unit when the nose image is captured, and thus even
the ROT from the
same subject animal may be varied.
In the approximation curve unit, the two approximation curves may be obtained
so that the
line segment connecting the center points of the approximation curves passes
the vertex points of
the two approximation curves when they are both approximated by ellipses.
By assuming that two nostril boundary curves are symmetric when they are
captured
directly from the front, the line segment connecting the two center points of
the two elliptical
nostril boundary curves should pass the vertex point of each ellipse. Using
this fact, the boundary
curves can be approximated by ellipses so that the line segment connecting the
center points of
the ellipses passes the vertex points of the ellipses.
A detailed account of the standardized ROT fixing unit is given below.
The standardized ROI fixing unit takes care of the transformation process of
the ROI
acquired by the ROT fixing unit into the standardized ROI when it is
necessary. FIG. 46 is a
diagram illustrating how the ROT from the same nose pattern image may be
varied depending on
33
CA 02925275 2016-03-23
the approximation curves of the nostril boundaries. As shown in FIG. 46, the
quadrilateral ROT
from even the same nose pattern image may be varied when different
approximation curves are
used, and the above quadrilateral ROI from even the same subject animal may
also be varied
depending on the relative position of the nose to the image acquisition unit
during capture.
To increase the identification rate, it is necessary to transform the given
ROI into the
standardized shape independent of the relative position of the nose and the
approximation curve
shapes. The standardized ROI fixing unit takes care of the transformation
process of the ROI into
the standard rectangular shape based on Equation (2).
FIG. 47 is a diagram illustrating the transformation process of the previously
determined
ROT into a standardized ROI as one embodiment of the present invention. As
shown in FIG. 47,
a quadrilateral ROT with four vertices 0, A, B, C is transformed into a
rectangular area of width
W and height H by Equation 2.
a a
P= 0 + ¨w (A ¨ 0), Q = B + ¨w (C ¨ B), R = 0 + ¨b (B ¨ 0), S = A + ¨b (C ¨ A)
H H
a
X = P + ¨b (Q ¨ P) = R + ¨w (S ¨ R) -------- (Equation 2)
H
In this transformation, the coordinates of the corresponding point X in the
ROT may not be
integral values in contrast to the point in the standardized ROT which has
integral coordinates a
and b. In this case, the brightness value K(X) of point X can be interpolated
from the nearby
pixel values and the intensity of the point (a, b) is defined by Equation 3.
I(a,b) = K(X) ---------- (Equation 3)
By the above transformation, various shapes of the quadrilateral ROT can be
transformed
into a rectangular shape.
34
CA 02925275 2016-03-23
In the present invention, except when necessary to distinguish between the ROT
and the
standardized ROT, both will be commonly referred to as ROIs for the sake of
simplicity. A
detailed account of the nose pattern code generation unit is given below.
FIG. 48 is a simplified block diagram describing the nose pattern code
generation unit as
one embodiment of the present invention. As shown in FIG. 48, the nose pattern
code generation
unit may comprise the frequency transform code generation unit 4801 and the
masking code
generation unit 4802.
A nose pattern code consists of a frequency transform code and a masking code
which are
generated by the frequency transform code generation unit and the masking code
generation unit,
respectively. A detailed account of the nose pattern code generation method is
given below.
FIG. 49 is a block diagram illustrating the process of generating nose pattern
codes from the
ROT. As shown in FIG. 49, a nose pattern code consists of the frequency
transform code
generated in the frequency transform code generation unit and the masking code
generated in the
masking code generation unit using the ROT. The nose pattern code is a 2-bit
array and its
component value is determined by predetermined frequency transform methods and
parameters
of the transforms.
The predetermined frequency transform methods may include several frequency
methods
including Gabor transform, Haar transform, Gabor Cosine transform, Gabor Sine
transform, Sine
transform, Cosine transform, and various wavelet transforms.
In the present invention, different frequencies for real and imaginary parts
of Gabor
transform may be used. Also, either of the real part of Gabor transform (Gabor
Cosine transform)
or the imaginary part of Gabor transform (Gabor Sine transform) may be used
alone. The choice
of frequency transform methods in the nose pattern code generation unit may be
determined
according to the performance and the processing speed of the image recognition
unit.
CA 02925275 2016-03-23
Below is a detailed account of the process of generating frequency transform
codes from the
ROT in the frequency transform code generation unit.
FIG. 50 is a diagram illustrating how to divide the ROT into smaller regions
with specified
dimensions which are called cell blocks and how frequency transform codes are
generated from
those cell blocks. Each cell block may consist of one or more pixels. When the
size of the given
nose image is large, a group of pixels may be reduced into one pixel by
averaging the values in
the group. The group of pixels may be regarded as a cell block in this
process. In other words,
each cell block may be represented by one pixel value using proper methods.
As shown in FIG. 50, the ROT with width W and height H is divided into n(=M*L)
equally
sized cell blocks. The total number of cell blocks and the size of each cell
block may be varied
depending on the size of the nose image, the breed of the subject animal, the
frequency transform
methods, parameters used in the frequency transform methods, etc.
Also the frequency transform codes consist of frequency transform values, each
of which is
obtained from a group of cell blocks called a cell-group as shown in FIG. 50.
In other words, a
cell-group is the basic unit for obtaining frequency transform codes. Two
different cell-groups
may include some common cell blocks.
The number of cell-groups and the number of cell blocks in each cell-group may
be varied
depending on the breed of the subject animal, frequency transform methods,
parameters used in
the frequency transform methods, etc. Each frequency transform value from a
cell-group is a
binary bit value (0 or 1) calculated based on the predetermined frequency
transform method and
parameters.
While, in most cases, each cell-group gives only one frequency transform value
so that the
length of the frequency transform code is equal to the number of cell-groups,
multiple frequency
transform values may be obtained from one cell-group with multiple frequency
transform
methods and parameters. Also, with some frequency transform method, multiple
frequency
transform values may be obtained from each cell-group even with one parameter.
36
CA 02925275 2016-03-23
For example, by Gabor transform, two frequency transform values - the real and
imaginary
parts of the frequency transform - are generated from each cell-group. While
frequency
transform methods vary in the way the binary bit value from each cell-group is
obtained, they are
essentially the same except their transformation formula and parameters. Thus,
the
transformation formula and its parameters are explicitly considered only of
Gabor transform,
Gabor Cosine transform and Gabor Sine transform in the present invention.
Below is a detailed account of the process of generating frequency transform
codes from the
ROT described above in the case of Gabor transform, Gabor Cosine transform and
Gabor Sine
transform.
[Example] Gabor transform, Gabor Cosine transform, Gabor Sine transform
The frequency transformation method of Gabor transform is given by Equation
(4) with its
parameters. Gabor Cosine transform and Gabor Sine transform each calculates
the binary bit
values of frequency transform codes using Equation (5) and Equation (6),
respectively.
wf H (a¨a0)2 ¨(1)-1,),0)2
V = 0 1(a, b)e-iwx(a-ao)e-i0 r
613-bo)e a2 e db da
--- (Equation 4)
W H (a_a0)2. __
Re(V) = 1-0 0 1(a, b)e a2 e cos(co,(a - a0) + wy(b - b0))db da
---- (Equation 5)
W H (a¨a0)2 (b¨b0)2
IM(V) = ¨ f 1(a, b)e 2 e 132 sin(cox(a - et0)+ coy(b - b0))db da
o o
-- (Equation 6)
37
CA 02925275 2016-03-23
In Equations (4), (5) and (6), I(a,b) denotes the brightness of the pixel in
the ROT at the
position of (a,b) and represent coordinates of a point in the ROT. Also, a, 13
are parameters of
Gabor transform to determine how large is the effective region to consider
when calculating the
frequency transform, and are parameters to determine the horizontal frequency
and the vertical
frequency, respectively.
Although the integration is done continuously in Equations (4), (5) and (6) in
the present
invention, the integration is approximated by the sum of integrand values at
finite lattice points.
Below is a detailed account of the process of approximating the integrations
in Equations
(4), (5) and (6). The region of integration in Equations (4), (5) and (6) is
different from the actual
region to consider when approximating the integration. FIG. 51 is a diagram
illustrating how the
two regions differ when using Gabor transform, Gabor Cosine transform, Gabor
Sine transform
as one embodiment of the present invention.
As shown in FIG. 51, the theoretical region of integration in Equations (4),
(5) and (6) is the
whole ROI, but the actual region to consider when approximating the
integration is restricted to
ca-a02 (b-b0)2
the region where the value of e a2
e )62 is greater than the given threshold since the
integration value has little difference with those pixels in the region where
the value of
(a-a0)2 (b-130)2
e a2 e
)32 is smaller than the given threshold. In this respect, cell-groups may be
formed
(a-a0)2 2b-b0>2
so that each cell-group only consists of cell blocks where the value of e
a2 e fl2 is
greater than the given threshold.
Such cell blocks are determined by the point (ao, 130) and parameters a, [3.
The region of
such a cell block is denoted as R(a0, 130, a,
below. When approximating the integration in
Equations (4), (5) and (6), only the region R(ao, 130, a, is
considered. In other words, the
actual region of integration in Equations (4), (5) and (6) is the region R
(a0, bo, a, [3). In FIG. 51,
the rectangle with dotted lines represents the cell-group, R (a0, 130, a, [3).
Thus, Equations (4), (5) and (6) can be approximated by Equations (4-1), (5-1)
and (6-1).
38
CA 02925275 2016-03-23
(a-a0)2 _(b-130)2
iw,(a-ao)e-iwy(b-bo)e a2 e )62 db da -------------
(Equation 4-1)
V = ffR (ao,b 0 ,a ,13 ) I (a' b) e-
(a-a0)2 (b-b0)2
Re(V)= f fR (a,,,b0 ,a,p) b)e
a2 e S2 cos (a ),(a ¨ a0)o) + wy (b ¨ 1)0)) db
da
I (a'
------ (Equation 5-1)
(a-a0)2 _03-1)02
IM (V) = ifI (a, b)e a2 e 1g2 sin (o),(a ¨ a0) + wy(b ¨ b0)) db da
R(a0,130,a,13)
------- (Equation 6-1)
When calculating Gabor Cosine transfoini or Gabor Sine transform, V in
Equation (4-1) is
evaluated on the region R(ao, 130, a, (3). In the case of Gabor Cosine
transform, the binary bit
value of the frequency transform code is determined by the sign of Re(V) in
Equation (5-1): it is
1 if Re(V)>O, and 0 otherwise. In the case of Gabor Sine transform, the binary
bit value of the
frequency transform code is determined by the sign of Im(V) in Equation (6-1)
in the same way.
In this way, for N configurations of frequency transform methods and values of
ao, 130, a,
13, wx, wy , N bit binary code is generated. Here, N denotes the length of the
frequency
transform code. Since multiple frequency transform values can be obtained from
each cell-group
using different frequency methods and their parameters, the length of the
frequency transform
code, N, may be different from the number of cell-groups.
Each binary value of the frequency transform codes can be obtained based on
its own
predetermined frequency transform method and parameters. In other words,
different frequency
methods or different parameters may be used for each binary value of the
frequency transform
code. In this way, the various features of a nose pattern image may be better
encoded into the
frequency transform code, and therefore increase the accuracy rate of
recognition in comparison
with the case where every binary value is calculated by using the same
frequency method or
parameters.
39
CA 02925275 2016-03-23
This strategy can also be applied to each breed type. Although the same
frequency
transform method or parameters may be chosen regardless of breed type, the
accuracy rate of
recognition may be increased if different frequency methods or parameters are
properly chosen
for each breed type.
For example, the best frequency transform method and parameters (e.g.,
frequencies) may
be chosen when generating nose pattern codes based on the estimated size of
nose pattern
features in the given nose pattern image. Using this strategy, a fine
distinction between breeds
may be achieved where the sizes of nose pattern features are significantly
different.
Below is a detailed account of the process of generating the masking code from
the ROT in
the masking code generation unit.
Each bit value of a masking code corresponds to a bit value of a frequency
transform code.
When a frequency code of N bits is generated from N configurations of
frequency transform
methods and values of ao, 130, a, p, wx, wy, each bit value of a masking code
is also computed
from each of the N configurations. Thus, the length of making codes is the
same as the length of
frequency transform codes.
The masking code generation process goes through a light-reflection masking
step and an
additional masking step. Depending on the methods of masking code generation,
both steps or
only one step may be applied.
(A) Light-Reflection Masking Step
When there are strong light reflections due to the wet nose or unexpected
illumination
problems, the nose pattern may appear damaged in the acquired nose image. Such
regions are
marked as damaged regions in the masking code.
More specifically, each value of the making code is assigned so that it can be
distinguished
from the value whether the corresponding frequency transform value is damaged
or not due to
the light reflections. For example, when the actual integration region R(a0,
bo, a, (3) includes
CA 02925275 2016-03-23
damaged regions due to light reflections as in FIG. 50, the value of 0 is
assigned to the
corresponding masking bit value to mark the frequency transform value as a
damaged one.
Otherwise, the value of 1 is assigned to the corresponding masking bit value.
(B) Additional Masking Step
The region containing nose patterns in the nose images of subject animals can
be damaged
due to nose or facial hairs, long whiskers, or foreign substances attached to
wet noses. When the
region R (a 0, 130, a, p) is damaged due to these reasons, the value of 0 is
assigned to the
corresponding masking bit value to mark the frequency transform value as a
damaged one.
Otherwise, the value of 1 is assigned to the corresponding masking bit value.
Once the nose pattern code consisting of the above-mentioned frequency
transform code
and masking code is generated, it is stored in the nose pattern code DB in the
image recognition
unit.
Below is a detailed account of the nose pattern code matching unit.
FIG. 52 is a simplified block diagram describing the nose pattern code
matching unit. As
shown in FIG. 52, the nose pattern code matching unit may include the nose
pattern code
verification unit 5201 and the nose pattern code identification unit 5202.
The nose pattern code verification unit performs verification (one-to-one
matching) by
comparing the nose pattern code generated for verification and the nose
pattern code stored in
the nose pattern code DB in the image recognition unit. Verification of the
generated nose
pattern code is performed by computing the dissimilarity (the distance) of two
nose pattern codes
using one of following matching methods: a) simple matching, b) shift matching
and c) block-
wise shift matching.
(A) Simple matching
FIG. 53 is a diagram illustrating the method of nose pattern code verification
through
simple matching. As shown in FIG. 53, the whole ROI A is compared to the whole
ROT B in the
41
CA 02925275 2016-03-23
simple matching. When the distance between the two nose pattern codes
corresponding to cell-
groups from Al to An of ROI A and cell-groups from B1 to Bn of ROI B is less
than the given
threshold, it is concluded that the two nose pattern images are taken from the
same subject.
Otherwise, it is concluded that the two images are taken from different
subject animals.
FIG. 54 is a diagram illustrating a situation in which different ROIs have
been selected from
the same nose pattern image of an individual for matching. As shown in FIG.
54, selecting the
same ROI is critical in increasing the accuracy rate of recognition since the
different ROIs from
the same subject animal result in high distance, and thus in a false non-match
in the simple
matching. There is a high probability of error in simple matching if it is
difficult to locate the
same ROI from the nose pattern images.
(B) Shift matching
As shown in FIG. 54, the probability of getting a false non-match is high when
ROI A and
ROI B are compared in simple matching even though they are from the same
subject animal.
More specifically, when cell-group Al from ROI A and cell-group B1 from ROI B
are compared
in simple matching, it is concluded that they are taken from different
individuals.
However, if cell-group Al from ROI A and cell-group B2 from ROI B are
compared, the
distance between the two cell-groups will be very small since they contain the
same nose pattern
and thus the same frequency transform codes. Therefore, there is a low
probability of error in
matching if partial regions (local regions) of two ROIs are compared rather
than the whole.
FIG. 55 is a diagram illustrating the method of nose pattern code
identification through shift
matching as one embodiment of the present invention. As shown in FIG. 55, the
local region a in
ROI A is compared to the local region b in ROI B. The nose pattern code
generated from cell-
groups through of the local region "a" is compared to the nose pattern code
generated from cell-
groups through of the local region "b".
In shift matching, the distance between two nose pattern codes is computed for
each pair of
local region "a" in ROI A and local region "b" in ROI B. By translating the
local regions a and b
42
CA 02925275 2016-03-23
in each ROT, multiple values of distance are computed. If the minimum of the
multiple values of
distance is less than the given threshold, it is concluded that the two nose
pattern images are
taken from the same individual. Otherwise, it is concluded that the two nose
pattern images are
taken from different subject animals.
For shift matching, nose pattern codes from all possible local regions in the
ROT should be
generated. In other words, the value of frequency transform should be computed
for each cell-
group in all possible local regions. Thus, it is required to compute all
values of the frequency
transform for all cell-groups in all possible local regions. For efficiency,
the values of frequency
transform from pre-computed cell-groups may be used rather than computing
every value of
frequency transform for every cell-group. When a cell-group is in a different
local region, the
pre-computed value from one local region is used for the other local region
rather than re-
computing the value of frequency transform for the cell-group. For this
reason, local regions and
cell-groups may be constructed with the efficiency of this computation in
mind. In the case
where this kind of computational efficiency is utilized, all values of
frequency transform from all
possible cell-groups are computed first and the nose pattern code for each
local region is
constructed using the pre-computed values of frequency transform.
FIG. 56 is a diagram illustrating a matching situation in which the ROIs of
the nose pattern
code selected from the same individual have nonidentical vertical and
horizontal proportions. As
shown in FIG. 56, it is critical to increasing the accuracy rate of
recognition that the
transformation from a quadrilateral ROT to a standard rectangular ROT should
be incorporated in
the shift matching since different shapes of ROI from the same subject animal
result in different
transformations, and in non-matching even though all of the translated local
regions are
compared. Therefore, the probability of error is high in the shift matching if
any of the translated
local regions of one ROT does not match any local region of the other ROT or
different
transformations are applied to make a standard rectangular ROT.
(C) Block-wise shift matching
As shown in FIG. 56, if the local region a in ROT A is compared to the
translated local
region of b in ROT B, the probability of recognizing the same source animal is
low. However,
43
CA 02925275 2016-03-23
there is a low probability of error in identification if a partial region
(called a slice region) of a
local region is selected and compared by translating the slice region in a
local region. For
example, as in FIG. 56, if slice regions A3, A5, A7 in local region "a- have
the same nose
pattern as slice regions B2, B5, B8 in local region "b," respectively, the
corresponding values of
frequency transform from those slice regions are the same. Thus, by
translating a slice region, it
is probable that a match will be made between ROI A and ROI B.
FIG. 57 is a diagram illustrating the method of nose pattern code
identification through
block-wise shift matching as one embodiment of the present invention. As shown
in FIG. 57,
slice region a of local region "a" in ROI A is compared to slice region p of
local region "b" in
ROI B. Local region "a" in ROI A and local region "b" in ROI B are subdivided
into n*m equal
sized pieces with n horizontal pieces and m vertical pieces. Then the distance
between the nose
pattern code corresponding to each piece of local region "a" and local region
"b" is computed.
In block-wise shift matching, slice regions a and [3 are translated in a given
range so that
multiple values of distance are computed. Thus, while only one distance is
computed for each
pair of local regions in shift matching, in block-wise shift matching the
distance is computed for
each pair of slice regions and thus multiple values of distance for each pair
of local regions. So,
one representative value of distance (called "final distance") from multiple
values of distance
needs to be computed to compare with the given threshold.
To compute the final distance for each pair of local regions in block-wise
shift matching,
the distance should be computed for each pair of slice regions. The distance
(called block
distance) between a pair of slice regions is the minimum value of all possible
distances computed
from all possible translations of slice regions. Then, the final distance may
be defined as one of
the minimum, a geometric average, an arithmetic average of all block
distances. When this final
distance is less than the given threshold, two animal subjects are regarded as
the same individual.
Otherwise, the two animal subjects are regarded as different individuals.
In these verification methods (simple matching, shift matching, and block-wise
shift
matching), the above ROI A can be the ROI from which a stored nose pattern
code is generated,
44
CA 02925275 2016-03-23
and the ROI B can be the one from which the nose pattern code for verification
is generated, and
vice versa.
Below is a detailed account of matching methods for Gabor Sine transform.
[Example] Gabor Sine Transform
(A) Simple matching
For a ROT, let C denote a frequency transform code from the ROI and M a
masking code
from the ROI generated by Gabor Sine transform. Then, the nose pattern code
for ROT A consists
of N bits of frequency transform code Cl and N bits of masking code Ml, and
the nose pattern
code for ROI B consists of N bits of frequency transform code C2 and N bits of
masking code
M2.
The distance (D) between two nose pattern codes can be computed by Equation
(7).
D = I(C1XOR C2) AND (Mi AND M2)1]
imi AND M21
-------------------- (Equation 7)
In Equation (7), XOR denotes the operator of bitwise Exclusive-OR, AND the
operator of
bitwise AND. The number of bits whose value is 1 in the array of bits A is
denoted by A.
When the above computed distance is less than the given threshold, it is
concluded that the
two animal subjects represented by those two nose pattern codes are identical,
and they are
different individuals otherwise.
(B) Shift matching
For a local region in a ROT, let C denote a frequency transform code from the
local region
and M a masking code from the local region generated by Gabor Sine transform.
Then, the nose
pattern code for local region "a" in ROT A consists of N bits of frequency
transform code Cl and
CA 02925275 2016-03-23
N bits of masking code Ml, and the nose pattern code for local region "b" in
ROI B consists of
N bits of frequency transform code C2 and N bits of masking code M2.
FIG. 58 is a diagram illustrating the process of nose pattern code
identification through shift
matching using Gabor Sine transform. Let RI be the set of all the points (6x,
63,) of the lower left
vertices of all the possible translated local regions in ROI A, and R2 the set
of all the points
(Ax, Ay) of the lower left vertices of all the possible translated local
regions in ROI B. FIG. 58 is
a diagram illustrating the case
where R1={(1,1)},
R2={(0,0),(1,0),(2,0),(0,1),(1,1),(2,1),(0,2),(1,2),(2,2)}.
Then, one nose pattern code (a(1,1)) from ROI A, and nine nose pattern codes
(b(0,0),
b(1,0), b(2,0), b(0,1), b(1,1), b(2,1),b(0,2), b(1,2), b(2,2)) from ROI B are
generated, and these
codes give a total of nine values of distance by one-by-one comparison. Thus,
in this case, it is
only necessary to see if the minimum of nine values of distance is less than
the given threshold.
In this way, when the nose pattern code for a local region "a" in ROI A is
denoted by
C1(6õ, 6y) , M1(6õ, 43y) and that for a local region -b" in ROI B denoted by
C2 (Ax, Ay),
M2(, Ay) the distance between the two ROIs can be computed by Equation (8).
1
D = min (C1 XOR C2) AND (M1AND M2)I
(8x,8y)ER1,(Ax4y)ER2 I M1 AND M21
--------------------------- (Equation 8)
In Equation (8), XOR denotes the operator of bitwise Exclusive-OR, AND the
operator of
bitwise AND. The number of bits whose value is 1 in the array of bits A is
denoted by A.
(C) Block-wise shift matching
For a slice region of a local region in a ROI, let C denote a frequency
transform code from
the slice region and M a masking code from the slice region generated by Gabor
Sine transform.
Then, the nose pattern code for slice region a of local region "a" in ROI A
consists of N bits of
frequency transform code Cl and N bits of masking code Ml, and the nose
pattern code for slice
46
CA 02925275 2016-03-23
region p of local region "b" in ROI B consists of N bits of frequency
transform code C2 and N
bits of masking code M2.
FIG. 59 is a diagram illustrating the process of nose pattern code
identification through
block-wise shift matching using Gabor Sine transform.
Let R1 be the set of all the points (6x, 6y) of the lower left vertices of all
the possible
translated regions from slice region a in ROI A, and R2 the set of all the
points (Ax, Ay) of the
lower left vertices of all the possible translated regions from slice region p
in ROI B. FIG. 58 is a
diagram illustrating the case where
R1=1(1,1)1,
R2= 1(0,0),(1,0),(2,0),(0, 1 ),( 1 , 1),(2,1 ),(0,2),(i ,2),(2,2)}.
Then, one nose pattern code (a(1,1)) from slice region a in ROI A, and nine
nose pattern
codes (p(o,o), 13(1,0), 13(2,0), 13(0,1), 13(1,1), 13(2,1), I3(0,2), PO ,2),
13(2,2)) from slice region 13 in
ROI B are generated, and these codes give a total of nine values of distance
by one-by-one
comparison. The minimum of all these values of distance is called the distance
between slice
regions, and the final distance between two ROIs may be defined as one of the
arithmetic
average, the geometric average, and the minimum of all possible distances
between two
corresponding slice regions.
In this way, when the nose pattern code for the translated region of the k-th
slice region a in
ROI A is denoted by Cl(k, 6õ, 6y), Ml(k, 6õ, Ely) and that for the translated
region of the k-th
slice region 13 in ROI B is denoted by C2(k, Ax, Ay), M2 (k, Ax, Ay), the
distance between two
corresponding k-th slice regions can be computed by Equation (9).
I (Ci XOR C2) AND (M1 AND M2)I
D (k) = min
(SSy)ER1,(A,,Ay)ER2 -- 1M1AND M21
----------------- (Equation 9)
In Equation (9), XOR denotes the operator of bitwise Exclusive-OR, AND the
operator of
bitwise AND. The number of bits whose value is 1 in the array of bits A is
denoted by A.
47
CA 02925275 2016-03-23
The final distance between two ROIs may be defined as one of the geometric
average, the
arithmetic average, and the minimum of all values of distance between two
corresponding slice
regions for all pairs of slice regions. When the above computed final distance
is less than the
given threshold, it is concluded that the two animal subjects represented by
those nose pattern
codes are identical, and they are different individuals otherwise.
In these verification methods (simple matching, shift matching, and block-wise
shift
matching), the above ROI A can be the ROT from which a stored nose pattern
code is generated,
and the ROT B can be the one from which the nose pattern code for verification
is generated, and
vice versa.
Below is a detailed account of the above nose pattern code identification
unit.
The nose pattern code identification unit performs identification (one-to-many
matching) by
comparing the nose pattern code generated for identification with multiple
nose pattern codes
stored in the nose pattern code DB in the image recognition unit.
While verification (one-to-one matching) requires computing the distance
between a single
nose pattern code and a single stored nose pattern code through simple
matching, shift matching
or block-wise shift matching, identification (one-to-many matching) requires
computing the
distances between a single nose pattern code and each of the multiple nose
pattern codes stored
in the DB.
FIG. 60 is a diagram illustrating the method of nose pattern code
identification (one-to-
many matching) as one embodiment of the present invention. As shown in FIG.
60, identification
is performed by computing the distances between a single nose pattern code
given for
identification and each of k nose pattern codes (nose pattern code_1, ,
nose pattern code_k)
stored in the DB. As previously described, each nose pattern code stored in
the DB consists of
the frequency transform code C and the masking code M, and the DB of pre-
registered nose
pattern codes may be constructed as a part of the image recognition unit.
48
CA 02925275 2016-03-23
From k nose pattern codes stored in the DB and a single nose pattern code
given for
identification, a total of k distances are computed, and these k distances may
be denoted by D1,
Dk. The values of Dl, Dk
may be sorted in a descending or ascending order. One or more
candidate nose pattern codes can be selected, according to one of three
predefined selection rules:
(a) a rule selecting the nose pattern code that gives the minimum distance
among all those that
yield distances less than the given threshold, (b) a rule selecting all the
nose pattern codes whose
distance is less than the given threshold, and (c) a rule selecting a
predefined number, say n, of
nose pattern codes whose distance is within top n least distances.
For example, as shown in FIG. 60, when the distances Dl, D2, Dk
are sorted in such a
way that D3 < D1 < Dk < D2 < ... and the values of D3, DI, Dk, D2 are all the
values less than
the given threshold, the group of selected nose pattern codes consists of only
one nose pattern
code whose distance is D3 by rule (a); all of D3, D1, Dk, D2 by rule (b); or
D3, D1 and Dk by
rule (c) with n=3.
The image recognition unit according to an embodiment of the present invention
as
described above may be implemented as a program and recorded on a medium
readable by a
computer, which includes all kinds of compatible data storing devices such as
ROM, RAM, CD
ROM, magnetic tapes, floppy disks, and optical data storage devices, and
carrier wave (e.g.,
transmission through the Internet).
The recording medium may also be distributed over network coupled computer
systems so
that the computer readable code can be stored and executed in a distributed
manner. In addition,
the functional programs, codes, and code segments for implementing the present
invention may
be easily construed by programmers skilled in the area relevant to the
invention.
The following looks at the technical configuration of the aforementioned
animal recognition
method involving the characteristics of the subject animal.
49
CA 02925275 2016-03-23
The flowchart in FIG. 11 summarizes the method of animal recognition in the
present
invention, starting with S1101 selecting and S1102 fitting the animal into the
appropriate body
stabilizer unit; S1103 fixing the nose of the subject onto the image
acquisition unit and S1104
acquiring the nose pattern image; S1105 at the image recognition unit,
generating a nose pattern
code using the raw or processed nose pattern image and S1106 enrolling and
identifying the
individual using the nose pattern code.
Further technical details have been omitted to avoid redundancy.
The sequence of operation for the body stabilizer unit as shown in FIG. 21 is
as follows:
S2101 select the appropriate body stabilizer unit for the subject animal by
taking into
consideration the overall size, leg length, feet size, head size, and the
relative location of the nose;
S2102 fit the subject animal into the upper body stabilizer unit; S2103 fasten
the upper body by
utilizing the upper body stabilizing brace subunit and upper body stabilizing
brace pressure
adjuster subunit to fit the shoulder width; S2104 fit the subject animal into
the lower body
stabilizer; S2105 fasten the lower body by utilizing the lower body
stabilizing brace subunit and
lower body stabilizing brace pressure adjuster subunit to fit the ankles or
legs; S2106 set the
stance width adjuster, and also the height adjuster to fit the subject's
height if necessary to
connect the upper and lower body stabilizer units; S2107 fasten the head by
utilizing the head
stabilizer unit, making sure to set the height adjuster unit to the correct
height and the horizontal
balance adjuster unit to have the nose facing the image acquisition unit head-
on.
Further technical details have been omitted to avoid redundancy.
An example of the image acquisition process is illustrated in FIG. 38; the
order of events
need not be limited as follows. S3801 the operator selects the species on the
display unit to start
the automatic mode, or chooses the manual mode; S3802 in automatic mode, once
the species
selection is made, pressing the capture button starts the acquisition of the
batch of nose pattern
images at n frames per second while the lens module is shifted about within
the preset range of
positions (adjusting values of a and b). In the automatic mode, the image
capture unit,
illumination unit and front unit are automatically adjusted to accommodate the
subject animal
CA 02925275 2016-03-23
based on the values stored in the parameter DB for FOV, focus, luminosity,
etc. However, in
manual mode, the operator visually evaluates the features and variables of the
nose pattern
images through the display unit, and selects the best image. S3803 The nose
pattern images
acquired from the sensor of the image capture unit are stored in the buffer,
upon which the main
processor unit calculates the individual scores and compares to the threshold
values in the
reference DB. S3804 Once the best image that passes all the thresholds is
selected, it is stored in
the buffer.
Further technical details have been omitted to avoid redundancy.
FIG. 40 illustrates the method by which an animal nose pattern image is
analyzed to be used
for identification, the order of which may be modified to better suit the
equipment or
circumstances: S4001 acquisition of the subject animal's nose pattern image by
utilizing the
body stabilizer unit and image acquisition unit; S4003 setting the ROT on the
(processed) nose
pattern image; S4005 generating a nose pattern code from the fixed ROI; S4006
enrolling the
generated nose pattern code; S4007 comparing the stored nose pattern code from
the enrollment
to the newly generated nose pattern code in one-to-one matching for
verification; and S4008
running one-to-many matching for identification. Images acquired from S4001
that have been
processed are called processed nose pattern images, and an additional step
S4002 for storing
them may be included. Also, the step S4004 that generates a standardized ROT
from the ROT
selected in S4003 may also need to occur.
Further technical details have been omitted to avoid redundancy.
The following steps describe the process of selecting usable nose pattern
images. The main
processor unit selects nose pattern images that are of sufficient quality to
be used in the image
recognition unit, out of all the images captured by the image capture unit.
When multiple nose
pattern images are obtained by the image capture unit, each image is given
individual scores on
specific variables, and images that pass the threshold set by the image
analysis unit are selected.
If none out of a particular batch meet the threshold, then that whole group is
discarded and a
request for a new batch is sent to the image capture unit. During this
selection process, the
51
CA 02925275 2016-03-23
images are evaluated on such criteria as, the amount of light reflection,
sharpness, contrast ratio,
ROT for capture, noise level, etc; and only those images that pass the
threshold for each variable
are accepted. When more than one image out of a single batch pass the
threshold the one with the
highest total score (sum of individual scores) is selected, and this process
may take place
simultaneously as the image acquisition in the image analysis unit or the
image recognition unit.
There are two types of variables: those that are not related to species-
specific characteristics
(Al-A3) and those that are (A4-Al2). The former includes sharpness Al
,contrast A2, and noise
level A3; the latter includes ROT for capture A4, presence of light reflection
A5, nostril location
A6, sharpness of nostril image A7, contrast level of nostril image A8, noise
level of nostril image
A9, sharpness of the border between the nostril and ROI A10, contrast level at
the border
between the nostril and ROI All, and noise level at the border between the
nostril and ROI Al2.
Variables may be appropriately added to or subtracted from the above list
depending on a subject
animal species' particular characteristics.
When selecting nose pattern images, each image in a batch is given individual
scores on
specific variables, and images that pass the threshold set by the image
analysis unit are selected.
When more than one image out of a single batch pass the threshold the one with
the highest total
score (sum of individual scores) is selected.
Further technical details have been omitted to avoid redundancy.
Below is a detailed account of the methods of processing raw nose pattern
images into
processed nose pattern images.
Freshly acquired nose pattern images may require processing in order to
increase the
identification rate. Raw acquired images may present different levels of noise
and blurring, and
may require contrast adjustments to normalize the distribution of pixel
values.
The present invention uses the histogram equalization technique to normalize
the
distribution of pixel values of images. In order to adjust the distribution of
pixel values, a
52
CA 02925275 2016-03-23
distribution function is fixed and applied with histogram equalization for
each nose pattern
image to have the same fixed distribution function.
Image filtering techniques may also be applied to take care of the noise and
blurring issues,
with Gaussian or median filters for noise level adjustment, and a variety of
low-pass filters in the
frequency domain. Moreover, sharpening techniques using derivatives can be
used to accentuate
the embossed nose patterns, and de-convolution techniques can be used to
restore damaged
images.
Below is a detailed account of the method of fixing the ROI from the
(processed) nose
pattern image.
In the first step, the nostril boundary needs to be found in order to fix the
ROI from the nose
pattern image.
FIG. 42 is a schematic diagram illustrating how to find the nostril boundary
as one
embodiment of the present invention. FIG. 42 illustrates the nostril boundary
setting process,
where the nostrils appear as a shade due to the indirect illumination. The
boundary of this shade
is the basis for the nostril boundary, which may take the form of a circular
or elliptical arc, etc. In
order to extract the boundary points, starting with a point(s) within the
shade as the center
point(s), the boundary points are located based on the change in brightness
along the ray from
the fixed center points. Points along the rays extending in various directions
that display a sharp
change in brightness are marked as candidate points, and the correct boundary
points are found
among those candidate points based on the statistical analysis of nostril
shape and location.
Using the above statistical analysis, not all of the boundary points of the
nostril in various
directions may be extracted resulting in that only parts of boundary points
are extracted.
Sometimes, even with indirect illumination certain areas that are not in the
nostrils may
appear to be inside of a similar shade, and therefore it is helpful to use
multiple center points and
53
CA 02925275 2016-03-23
statistical analysis utilizing the shape and location information of nostrils
to prevent finding
incorrect boundary points.
The nostril boundaries are approximated by curves fitting the boundary points
found in the
above step. In this approximation, the final approximation curve is the best
curve fitting the
boundary points found by various regression analyses, and it is usually a
circular arc or elliptical
arc.
Although the left and right nostril boundaries can be regarded as symmetric
curves when
they are seen from the front of the nose, the two approximation curves can be
asymmetric
ellipses if the nose pattern image is taken from askew.
Also, since the curve approximation unit separately approximates the left and
right nostril
boundaries, the two approximation curves can have different shapes resulting
in that one curve is
a circle, and the other an ellipse. It is also possible that the two
approximation curves are
different in size although they are all either circle or ellipses.
A quadrilateral region of a nose pattern image between the two approximation
curves
obtained from the above step is to be defined. This process consists of two
steps: a) the region
between two approximation curves is identified and b) a quadrilateral region
contained in the
identified region is extracted.
(A) The first step where the region between two approximation curves is
identified:
FIG. 44 is a schematic diagram illustrating how to identify the region between
the two
approximation curves (circles or ellipses) as one embodiment of the present
invention. As shown
in FIG. 44, two points which are on the intersections between each
approximation curve and the
line segment connecting two centers of the approximation curves are located,
and the two
tangent lines which tangent at each located point to the approximation curve
(the left tangent line
is denoted by T L, and the right tangent line by T R) are found. These tangent
lines may be
perpendicular to the line segment connecting the two centers when the two
approximation curves
are symmetrical, and may not be perpendicular when they are not symmetrical.
54
CA 02925275 2016-03-23
The two connecting lines are then found: one line connecting two upper vertex
points of the
approximation curves and the other line connecting two lower vertex points
(the upper line is
denoted by T U, and the lower line denoted by T_D ). In this step, the two
connecting lines are
tangent lines which tangent to the both of the approximation curves when they
are both circles
and the two lines connects two upper vertex points or two lower vertex points
when they are both
ellipses.
(B) The second step where the quadrilateral region between two approximation
curves is
extracted as the ROT.
FIG. 45 is a schematic diagram illustrating how to extract the quadrilateral
region between
the two approximation curves as one embodiment of the present invention. As
shown in FIG. 45,
the ROT is the quadrilateral region encompassed by four lines obtained in Step
A. The shape and
the size of the ROT may be varied depending on the relative position of the
nose to the position
of the image acquisition unit when the nose image is captured, and thus even
the ROI from the
same animal subject may be varied.
In the approximation curve unit, the two approximation curves may be obtained
so that the
line segment connecting two center points of the approximation curves passes
each of vertex
points of two approximation curves when they are both approximated by
ellipses.
By assuming that two nostril boundary curves are symmetric when they are
captured
directly from the front, the line segment connecting the two center points of
the two elliptical
nostril boundary curves should pass the vertex point of each ellipse. Using
this fact, the boundary
curves can be approximated by ellipses so that the line segment connecting two
center points of
ellipses passes the vertex points of ellipses.
Further technical details have been omitted to avoid redundancy.
Below is a detailed account of the step to generate the standardized ROT from
the above
ROI.
CA 02925275 2016-03-23
The ROI is transformed into the standardized ROI when it is necessary to
normalize the
above ROI obtained from the step of ROI fixing. FIG. 46 is a diagram
illustrating how the ROI
from the same nose pattern image may be varied depending on the approximation
curves of the
nostril boundaries. As shown in FIG. 46, the quadrilateral ROI from even the
same nose pattern
image may be varied when different approximation curves are used, and the
above quadrilateral
ROI from even the same subject animal may also be varied depending on the
relative position of
the nose to the image acquisition unit during capture.
To increase the identification rate, it is necessary to transform the given
ROI into the
standardized shape independent of the relative position of the nose and the
approximation curve
shapes. The standardized ROI fixing unit takes care of the transformation
process of the ROI into
the standard rectangular shape based on Equation (2).
By the above transformation various shapes of quadrilateral ROIs can be
transformed into a
rectangular shape, which is the standard ROI and thusly stored in the memory
or the DB. The
technical details of the standard ROI have been omitted to avoid redundancy.
Below is a detailed account of the process of generating nose pattern codes
from the above-
stated ROI.
Nose pattern codes are generated via steps of generating the a) frequency
transform code
and b) masking code from the ROI.
Below is a detailed account of the process of generating frequency transform
codes from the
ROI stated above.
FIG. 49 is a block diagram illustrating the process of generating nose pattern
codes from the
ROI. As shown in FIG. 49, a nose pattern code consists of the frequency
transform code and the
masking code generated from the whole ROI. The nose pattern code is a 2-bit
array and its
56
CA 02925275 2016-03-23
component value is determined by predetermined frequency transform methods and
parameters
of the transforms.
The predetermined frequency transform methods may include several frequency
methods
including Gabor transform, Haar transform, Gabor Cosine transform, Gabor Sine
transform, Sine
transform, Cosine transform, and various wavelet transforms.
In the present invention, different frequencies for real and imaginary parts
of Gabor
transform may be used. Also, either of the real part of Gabor transform (Gabor
Cosine transform)
or the imaginary part of Gabor transform (Gabor Sine transform) may be used
alone. The choice
of frequency transform methods may be determined based on the performance and
the
processing speed of the image recognition unit. The technical details
regarding the generation of
frequency transform codes have been omitted to avoid redundancy.
Below is a detailed account of the process of generating masking codes from
the ROI stated
above.
Each bit value of a masking code corresponds to a bit value of a frequency
transform code.
When a frequency code of N bits is generated from N configurations of
frequency transform
methods and values of ao, 130, a, 13, orx, ei3, each bit value of a masking
code is also computed
from each of the N configurations. Thus, the length of making codes is the
same as the length of
frequency transform codes.
The masking code generation process goes through a light-reflection masking
step and an
additional masking step. Depending on the methods of masking code generation,
both steps or
only one step may be applied. The technical details regarding the generation
of masking codes
have been omitted to avoid redundancy.
Below is a detailed account of the process of verification (one-to-one
matching), in which a
nose pattern code generated for verification is compared to a nose pattern
code stored in the nose
pattern code DB.
57
CA 02925275 2016-03-23
The nose pattern code verification unit performs verification (one-to-one
matching) by
comparing the nose pattern code generated for verification and the stored nose
pattern code.
Verification of the generated nose pattern code is performed by computing the
distance between
two nose pattern codes through one of the following matching methods: a)
simple matching, b)
shift matching and c) block-wise shift matching. The technical details
regarding the matching
methods have been omitted to avoid redundancy.
Below is a detailed account of the process of identification (one-to-many
matching), in
which the distances between the nose pattern code generated for identification
and each of the
multiple nose pattern codes stored in the nose pattern code DB is computed.
While verification (one-to-one matching) requires computing the distance
between a single
nose pattern code and a single stored nose pattern code through simple
matching, shift matching
or block-wise shift matching, identification (one-to-many matching) requires
computing the
distances between a single nose pattern code and each of the multiple nose
pattern codes stored
in the DB.
FIG. 60 is a diagram illustrating the method of nose pattern code
identification (one-to-
many matching) as one embodiment of the present invention. As shown in FIG.
60, identification
is performed by computing the distances between a single nose pattern code
given for
identification and each of k nose pattern codes (nose pattern code_1, ,
nose pattern code_k)
stored in the DB. As previously described, each nose pattern code stored in
the DB consists of
the frequency transform code C and the masking code M, and the DB of pre-
registered nose
pattern codes may be constructed as a part of the image recognition unit.
From k nose pattern codes stored in the DB and a single nose pattern code
given for
identification, a total of k distances are computed, and these k distances may
be denoted by DI,
Dk. The values of D1, Dk
may be sorted in a descending or ascending order. One or more
candidate nose pattern codes can be selected, according to one of three
predefined selection rules:
(a) a rule selecting the nose pattern code that gives the minimum distance
among all those that
58
CA 02925275 2016-03-23
yield distances less than the given threshold, (b) a rule selecting all the
nose pattern codes whose
distance is less than the given threshold, and (c) a rule selecting a
predefined number, say n, of
nose pattern codes whose distance is within top n least distances.
Further technical details have been omitted to avoid redundancy.
Finally, it should be noted, that certain components of the present invention
have been
described as operating in combination, or combined into one, does not suggest
that the invention
is necessarily limited to such embodiments. In other words, within the scope
of the present
invention, such components may operate by selectively binding to one or all of
the other
components. Moreover, although each component may be implemented as an
independent
hardware, a part of or all of each component may be optionally combined to
perform one or a
combination of some or all functions of...
In addition, the functional programs, codes, and code segments for
implementing the
present invention may be easily construed by programmers skilled in the area
relevant to the
invention. Such a computer program may be saved onto a computer readable media
to be read
and executed by a computer, and possible storage devices include ROM, RAM, CD
ROM,
magnetic tapes, floppy disks, and optical data storage devices, and carrier
wave.
Moreover, terms such as "to include," "to comprise" or "to have" as set forth
above, unless
specified otherwise, should not be interpreted to exclude other components,
but rather that it may
also include others.
All terms, including technical and scientific terms, unless defined otherwise,
should be
interpreted to mean the conventional definition. Other commonly used terms
should be
understood to match the contextual meaning relevant to the art.
59
CA 02925275 2016-03-23
Industrial Applicability
The present invention relates to the apparatus and method of animal
recognition using nose
patterns. Specifically, it involves the use of a body stabilizer unit, image
acquisition unit, and
image recognition unit to obtain identifiable nose pattern images. The
proposed method is
physically and economically easy to implement and requires little expertise on
the part of the
operator; and as such, presents a very high potential for industrial
applicability.